Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 03143933 2021-12-16
WO 2021/216135 PCT/US2020/066932
- 1 -
ENHANCING SIGNATURE WORD DETECTION IN VOICE ASSISTANTS
Background
[0001] The present disclosure relates to speech recognition systems and, more
particularly,
to systems and methods related to speech-assisted devices with signature word
recognition.
Summary
[0002] Smart voice-assisted devices, smart devices commanded to perform
certain tasks, are
now ubiquitous to modern households and the commercial sector. The utterance
of a
signature word or phrase signals the device of a command or a query intended
for the device
to perform. The phrase "Ok, Google, play Game of Thrones!", when spoken
clearly into a
Google-manufactured voice-assisted system, is commonly known to cause the
device to carry
out the user command to play the television series "Game of Thrones" on a
media player, for
example. Similarly, uttering "Alexa, please tell me the time!" causes a
properly configured
speech-recognition device, such as the Amazon Echo, to announce the current
time. Both
"Ok, Google" and "Alexa", spoken within an acceptable range of a corresponding
properly
configured device, trigger a device reaction. But in the absence of a
signature word, and
particularly a signature word that precedes each user command or query, the
device fails to
take the commanded action and instead performs no response. The voice-assisted
device is
effectively deaf to a user command without a preceding signature word. The
signature word
is therefore key to the operation of voice-assisted devices. What is perhaps
even more key to
the proper operation of such devices is the order in which the signature word
appears in the
spoken command or query. That is, what grabs the attention of a smart voice-
assisted device
to carry out a user-voiced command, e.g., "Play Game of Thrones" or "Please
tell me the
time," is not only a signature word but also the utterance of the signature
word in a predefined
order, immediately before the spoken command, a structured and rather rigid
approach to
proper processing of a user command.
[0003] Repeating a signature word before uttering a command or query may seem
somewhat burdensome or unnatural for some users. It is rather atypical, for
instance, for a
friend to call a person by their name each time before uttering a sentence
directed to the
friend. "Jack, please stop watching tv," followed by "Jack, please get my bag
from the table,"
CA 03143933 2021-12-16
WO 2021/216135 - 2 - PCT/US2020/066932
followed by "Jack, let's go" sounds awkward and unusual. Speaking a signature
word in the
beginning, middle or the end of a query or command should serve no
consequence, yet, in
today's devices, it does.
[0004] It is no secret that voice-assisted devices raise privacy concerns by
capturing vast
amounts of recognizable and private communication spoken within a speaking
range of the
device. Long before a signature word, such as "Ok, Google," "Alexa" or "TIVO,"
is detected,
all surrounding conversations are locally or remotely recorded. Moreover,
certain privacy
regulations remain unaddressed. Absent proper user consent, an entire
household of speech
and conversation, over a span of numerous days, weeks, months, and in many
cases years, are
unnecessarily and intrusively recorded and made available to a remotely
located device
manufacturer, completely removed from user control. Worse yet, many users
remain ignorant
of voice-assisted data collection privacy violations. Recent privacy law
enactments, in
Europe, California, and Brazil, for example, demand manufacturers to place
privacy rights of
their users front and center by requiring express user consent before user
data collection, a
condition not readily met by current-day smart devices.
[0005] Accordingly, a less stringent and less intrusive electronic voice
assistant device, one
without a strict pre-command signature word requirement and with a more
natural user
communication protocol, would better serve a voice-assistant user. In
accordance with
various speech recognition embodiments and methods disclosed herein, a user
event
indicative of a user intention to interact with a speech recognition device is
detected. In
response to detecting the user event, an active mode of the speech recognition
device is
enabled to record speech data based on an audio signal captured at the speech
recognition
device irrespective of whether the speech data comprises a signature word.
While the active
mode is enabled, a recording of the speech data is generated, and the
signature word is
detected in a portion of the speech data other than a beginning portion of the
speech data. In
response to detecting the signature word, the recording of the speech data is
processed to
recognize a user-uttered phrase.
[0006] In some embodiments, a method of detecting a sentence includes at least
one of a
command and a query in a speech recognition system. Speech data is buffered
based on an
audio signal captured at a computing device operating in an active mode. The
speech data is
buffered irrespective of whether the speech data comprises a signature word.
The buffered
speech data is processed to detect the presence of a sentence comprising at
least one command
and the query for the computing device. Processing the buffered speech data
includes
CA 03143933 2021-12-16
WO 2021/216135 - 3 - PCT/US2020/066932
detecting the signature word in the buffered speech data, and, in response to
detecting the
signature word in the speech data, initiating detection of the sentence in the
buffered speech
data.
Brief Description of the Drawings:
[0007] The above and other objects and advantages of the disclosure will be
apparent upon
consideration of the following detailed description, taken in conjunction with
the
accompanying drawings, in which:
[0008] FIGS. 1-2 each show an illustrative block diagram of a distinct speech
recognition
system, in accordance with some embodiments of the disclosure;
[0009] FIG. 3 depicts an illustrative flowchart of a speech recognition
process, in accordance
with some embodiments of the disclosure;
[0010] FIG. 4 depicts an example speech detection technique, in accordance
with some
embodiments of the disclosure;
[0011] FIG. 5 depicts an illustrative flowchart of a speech recognition
process, in accordance
with some embodiments of the disclosure;
[0012] FIG. 6 depicts an illustrative flowchart of a speech recognition
process, in accordance
with some embodiments of the disclosure;
[0013] FIG. 7 is a block diagram of an illustrative user device, in accordance
with some
embodiments of the present disclosure; and
[0014] FIG. 8 is a block diagram of an illustrative system for transmitting
information, in
accordance with some embodiments of the present disclosure.
Detailed Description
[0015] FIG. 1 shows an illustrative block diagram of speech recognition system
100, in
accordance with some embodiments of the present disclosure. System 100 is
shown to
include a speech recognition device 102 communicatively coupled to a
communication
network 104, in accordance with various disclosed embodiments. Speech
recognition
device 102 is shown to include an active mode buffer 116, a user activity
detector 118 and an
audio signal receiver 120. Communication network 104 is shown to include a
speech
recognition processor 124. In some embodiments, speech recognition device 102
may be
implemented, in part or in whole, in hardware, software, or a combination of
hardware and
software. For example, a processor (e.g., control circuitry 704 of FIG. 7)
executing program
code stored in a storage location, such as storage 708 of FIG. 7, may perform,
in part or in
CA 03143933 2021-12-16
WO 2021/216135 - 4 - PCT/US2020/066932
whole, some of the speech recognition functions of device 102 disclosed
herein. Similarly,
speech recognition processor 124 may be implemented, in part or in whole, in
hardware,
software, or a combination of hardware and software. For example, a processor
(e.g., control
circuitry 704 of FIG. 7) executing program code stored in a storage location,
such as
storage 708 of FIG. 7, may perform, in part or in whole, some of the speech
recognition
functions of processor 124 disclosed herein.
[0016] Communication network 104 may be a wide area network (WAN), a local
area
network (LAN), or any other suitable network system. Communication network 104
may be
made of one or multiple network systems. In some embodiments, communication
network 104 and device 102 are communicatively coupled by one or more network
communication interfaces. In some example systems, communication network 104
and
device 102 are communicatively coupled by the interfaces shown and discussed
relative to
FIG. 7. Communication network 104 and device 102 may be communicatively
coupled in
accordance with one or more suitable network communication interfaces.
[0017] In accordance with an embodiment, speech recognition device 102
receives audio
signals at audio signal receiver 120, processes the received audio signals
locally for speech
recognition, and transmits the processed audio signals to communication
network 104 for
further speech recognition processing. For example, speech recognition device
102 may
receive audio signals 110 and 114 from each of users 106 and 112,
respectively, process the
received signals 110 and 114 for speech processing with user activity detector
118 and active
mode buffer 116 and transmit the processed audio signals to speech recognition
processor 124
of communication network 104 for further voice recognition processing. In some
embodiments, processor 124 transmits the processed speech file to a third-
party transcription
service for automated speech recognition to translate voice into text and
receive a text file
corresponding to the transmitted processed speech file. For example, processor
124 may send
the processed speech file to Amazon Transcribe and Google Speech-to-Text.
[0018] In some embodiments, user activity detector 118 includes detecting and
sensing
components sensitive to recognizing a physical change related to the user,
such as, but
without limitation, a physical user movement closer in proximity to speech
recognition
device 102. For example, user 106 may make a sudden physical head turn from a
starting
position 106a not directly facing the audio signal receiver 120 of device 102,
to a turned
position 106b, directly facing the audio signal receiver 120 of device 102. To
user activity
detector 118, the detected user 106 turn action signals a soon-to-follow audio
signal 110 with
a command or an assertion speech originating from user 106 or from the
direction of user 106.
CA 03143933 2021-12-16
WO 2021/216135 - 5 - PCT/US2020/066932
In contrast, in the absence of a physical change in user 112, activity
detector 118 detects no
user activity, user movement or audio strength change, from user 112 or from
the direction of
user 112 that may suggest user 112 is possibly interested in interacting with
device 102.
[0019] User activity detector 118 may detect a user event in a variety of
ways. For
example, user activity detector 118 may implement a motion detection function,
using a
motion detector device, to sense user 106 turn motion from position 106a to
position 106b.
Activity detector 118 may alternatively or in combination implement a spectral
analysis
technique, using a spectral analyzer device, to detect an increased audio
signal amplitude
when receiving audio signal 110, corresponding to user 106, as user 106 turns
from
position 106a to position 106b, directly facing audio signal receiver 120 of
device 102. Still
alternatively or in combination, activity detector 118 may implement an image
capturing
function, using an image capturing device such as, without limitation, a
digital camera, that
captures images showing the user 106 turn movement from position 106a to
position 106b.
Device 102 may employ any suitable technique using a corresponding suitable
component
that helps detect a closer proximity of user 106 to device 102. In the non-
active mode where
device 102 is waiting to detect a user movement, such as discussed above,
device 102 remains
in a continuous intimation detection mode with functionality limited, in large
part, to the
detection with a reduced power consumption requirement. In response to a
detected user
activity, device 102 enables an active mode.
[0020] In the active mode, device 102 may start to record incoming audio
signals, such as
signal 110, in a storage location, such as storage 708 (FIG. 7). Audio signal
110 is made of
audio/speech chunks, packets of speech data. In some embodiments, device 102
saves the
speech data packets in the active mode, in active mode buffer 116. Buffer 116
may be a part
of or incorporated in storage 708 (FIG. 7). Audio signal receiver 120 may be a
microphone
internally or externally located relative to device 102.
[0021] In accordance with an example operational application, device 102 is a
TIVO voice-
enabled product. As depicted in FIG. 1, at 1), user activity detector 118
senses the user 106
turn movement from position 106a to position 106b and, in response to
detecting the user
turn, device 102 enables its active mode. While in the active mode, device 102
starts to
record incoming user utterances in the form of packets of speech data and, at
2), looks for a
signature word in the incoming speech data packets. Device 102 stores the
incoming speech
data packets in active mode buffer 116, a local storage location. At 3), in
response to
detecting the signature word, for example, signature word "TIVO," in a user
106 utterance,
i.e., "Please tell me the time, TIVO!", device 102 begins a processing phase
by transmitting
CA 03143933 2021-12-16
WO 2021/216135 - 6 - PCT/US2020/066932
the recorded speech data packets, in the form of an audio file, from buffer
116 to
communication network 104. Detection of the signature word, "TIVO," at 3) in
FIG. 1,
effectively starts the processing of the received speech data packets. At
communication
network 104, the transmitted packets are processed to recognize the user
utterance "Please tell
me the time, TIVO!", as shown at 4) in FIG. 1. As used herein, the term
"signature word"
refers to a word, phrase, sentence, or any other form of utterance that
addresses a smart
assistance device.
[0022] In some embodiments, recording, prompted by a user activity as
discussed above,
continues even after transmission and processing of the packets begins at
communication
network 104. In some embodiments, recording stops in response to packet
transmission to
and processing by communication network 104.
[0023] As earlier noted, device 102 records user 106 utterances locally
without sharing the
recorded information with communication network 104 for privacy reasons. User
speech is
therefore maintained confidentially until a signature word detection. In the
case where no
signature word is detected, no recording of user utterances is generated. In
some
embodiments, in furtherance of user privacy protection, prior to starting to
generate a
recording, device 102 may request a privacy consent (e.g., consent to the
collection of user
speech) confirmation from user 106 and may further condition the recording on
receiving the
consent. That is, device 102 simply does not record user utterances even in
the presence of a
signature word detection unless a user consent acknowledgement is received.
For example,
device 102 may generate a display on a user device, such as a user smartphone
or a user
tablet, with privacy terms to be agreed to by the user. Device 102 may wait to
receive a
response from the user acknowledging consent to the terms by, for example,
clicking a
corresponding box shown on the user device display.
[0024] In some embodiments, device 102 encrypts speech data packets
corresponding to
user 106 utterances, for example, utterance "Please tell me the time, TIVO!",
before storing or
recording the packets in buffer 116, as yet another added security measure to
ensure meeting
stringent legal privacy requirements.
[0025] In accordance with some embodiments, the signature word, "TIVO," is
detected
despite its location in the user-uttered phrase. "TIVO" may appear in the
beginning, middle,
end, or anywhere in between, in the phrase "Please tell me the time" yet be
recognized in
accordance with some disclosed embodiments and methods. For example, the user
106 turn
(from 106a to 106b) sets off a recording session guaranteeing preservation of
the signature
word despite the signature word location in the phrase.
CA 03143933 2021-12-16
WO 2021/216135 - 7 - PCT/US2020/066932
[0026] As previously indicated, the speech data packets may be saved in a
single and local
physical buffer with no other storage location necessitated, in part, because
pre-active mode
recording is unnecessary. This single buffer approach is yet another effective
device 102
energy-conservation measure.
.. [0027] FIG. 2 shows an illustrative block diagram of speech recognition
system 200, in
accordance with some embodiments of the present disclosure. In an example
embodiment, as
discussed below, system 200 is configured as system 100 of FIG. 1 with further
processing
features shown and discussed relative to FIG. 2.
[0028] System 200 is shown to include a speech recognition device 202
communicatively
coupled with a communication network 204. With continued reference to the
operational
example of FIG. 1, in FIG. 2, an activity detector 218 of device 202 detects a
turn motion
from position 206a to position 206b by user 206 and, in response to the
detection, device 202
enables the active mode. In the active mode, device 202 records incoming
speech data
packets corresponding to the user utterance "Please tell me the time, TIVO !",
in active mode
buffer 216. Analogous to the example of FIG. 1, in FIG. 2, device 102 stores
at least speech
data packets corresponding to three phrases 234, namely phrases 1, 2, and 3
(234a, 234b, and
234c), originating from user 206, in buffer 216. The phrases are stored in an
audio file 230 in
buffer 216. Audio buffer 230 may have a different number of phrases than that
shown and
discussed herein.
.. [0029] Audio file 230 further includes silent durations 232, each of which
(silent
duration 232a, silent duration 232b, and silent duration 232c) is located
between two adjacent
phrases in audio file 230. In some embodiments, device 102 performs some or
all audio file
processing locally. For example, device 102 may perform detection and
recognition of a
sentence, as disclosed herein, locally. In some embodiments, device 102 and a
speech
recognition processor 224 of communication network 204 share the tasks. In yet
another
embodiment, device 202 transmits audio file 230 to communication network 204
for
processing by processor 224, as discussed in large part relative to FIG. 1.
The discussion of
FIG. 2 to follow presumes the last scenario with device 202 transmitting audio
file 230 for
processing by communication network 204.
[0030] In some embodiments, device 202 transmits audio file 230 to
communication
network 204 as buffer 216 becomes full, on a rolling basis. In this
connection, in accordance
with some embodiments, buffer 216 is presumed adequately large to accommodate
at least a
phrase worth of speech data packets. In some embodiments, device 202 transmits
less than a
buffer full of phrases to communication network 204. For instance, device 202
may transmit
CA 03143933 2021-12-16
WO 2021/216135 - 8 - PCT/US2020/066932
one, two, or three phrases as they become available in buffer 216 to
communication
network 204. In this scenario, device 202 is equipped with the capability to
detect the
beginning and ending of a phrase. In some embodiments, device 202 may detect
silent
durations 232 to attempt to distinguish or parse a sentence.
[0031] In some embodiments, as speech data packets are received at an audio
signal
receiver 220 of device 202, device 202 may implement or solicit a speech
detection algorithm
to determine the start and end of a phrase based on a sequence validating
technique. For
example, device 202 may implement a segmental conditional random field (CRF)
algorithm
or use a hidden Markov model (HMM) or a long short-term memory (LSTM) model to
predict the end of the audio signal corresponding to a phrase or sentence (or
the beginning of
a silent duration 232 in FIG. 2). In implementations using model-based
prediction, such as
with the use of HMM or LSTM models, the model is trained to predict whether
the uttered
word is a start of the sentence, an intermediate word or the last word of the
sentence. As
further described relative to FIG. 4, a model is trained with and can
therefore predict features
such as, without limitation, question tags, WH ("what") words, articles, part-
of-speech tags,
intonations, syllables, or any other suitable language attributes. The term
"tag," as used
herein, refers to a label that is attached to, stored with, or otherwise
associated with a word or
a phrase. For instance, "verb" is an example of a part-of-speech tag that may
be associated
with the word "running." As used herein, the term "feature" refers to a
collection of different
types of tag values. Part-of-speech is one example of a feature or a type of
tag value. An
influential word is another example of a feature or a type of tag value.
During the training of
the model, a collection of word-to-tag mappings is fed to the model along with
an input
sentence. As used herein, the term "label" refers to a value or outcome that
corresponds to a
sample input (e.g., a query, features, or the like) and that may be employed
during training of
the model. In some examples, the model is trained by way of supervised
learning based on
labeled data, such as sample inputs and corresponding labels. In some
examples, features
may be referred to as dependent variables, and labels may be referred to as
independent
variables.
[0032] A sequence validation technique may be executed on a sentence or phrase
in a
forward and a backward direction for improved prediction reliability but at
the expense of
requiring a separate model and model training for each direction, a rather
costly approach. A
sequence structure validation may be employed using conditional probability at
its base, for
example, the Bayes theorem, to store states at different points in time of a
sentence. In some
embodiments, an extension to the basic sequence structure validation algorithm
may be
CA 03143933 2021-12-16
WO 2021/216135 - 9 - PCT/US2020/066932
implemented with Markov chains. Markov chains introduce hidden states at every
state
transition, for example, between the words of a phrase or sentence, or between
syllables of
words of a phrase or sentence. The labels used for each such training example
are the points
in time at which the phrase (spoken utterance) may start and end.
[0033] In some embodiments, the start of a phrase is typically driven by
decisions taken
during the handling of the last packet of a phrase, and a list of contextual
information is
passed to the next audio chunk (or packet). In some cases, a silent duration
of a predefined
duration may be detected in real time to help shift to a new context. In some
embodiments,
silent duration detection may be implemented based on heuristics. For example,
heuristics of
reconfigurable manufacturing systems (RMS) values representing speech data
amplitude may
be processed to detect silent durations in an audio file, such as the audio
file 230 of FIG. 2.
[0034] In implementations with communication network 204 facilitating packet
processing,
processor 224 may achieve phrase detection by implementing the foregoing
speech detection
algorithms described with reference to device 202. For example, in an instance
of audio
file 230, audio file 230', shown at processor 224 of communication network 204
in FIG. 2,
silent duration 232' (232a', 232b', and 232c') may be detected to isolate or
distinguish each
of the phrases 234' (234a', 234b', and 234c'). In the example of FIG. 2,
phrase 2, 234b' is
shown detected at processor 224.
[0035] FIG. 3 shows an illustrative flowchart of a speech recognition process
300, in
accordance with some embodiments of the disclosure. Process 300 may be
performed, partially
or in its entirety, by a voice-assisted device, such as devices 102 and 202 of
FIGS. 1 and 2,
respectively. In some embodiments, process 300 may be performed by control
circuitry 704
(FIG. 7). In some embodiments, process 300 may be performed locally or
remotely or a
combination thereof For example, process 300 may be performed, partially or in
its entirety,
by processor 124 or processor 224 of FIGS. 1 and 2, respectively. Process 300
may be
performed by a combination of a voice-assisted device and a remote process,
for example,
device 102 and processor 124 or device 202 and processor 224.
[0036] At 302, process 300 begins, and at step 304, a device implementing
process 300 waits
for the detection of a user event, such as a user movement, as previously
discussed. In response
to the detection of a user event at step 304, process 300 proceeds to step
306, and an active
mode of the device is enabled to start generating a recording of the incoming
speech data
packets. Next, at step 308, the speech data is recorded and process 300
proceeds to step 310.
At step 310, the device implementing process 300 looks for a signature word in
the recorded
speech data. In response to the detection of a signature word at step 310,
process 300 proceeds
CA 03143933 2021-12-16
WO 2021/216135 - 10 - PCT/US2020/066932
to step 312, and at step 312, the recorded speech data is processed as
described in accordance
with various disclosed methods. For example, the recorded speech data may be
transmitted to a
network cloud device for processing. After step 312, process 300 resumes
starting at step 304
to look for the next user event. At step 304, a device implementing process
300 waits to detect
a user event before proceeding to step 306, and in some embodiments, the
device may abandon
waiting for detection in response to a time out period or in response to a
manual intervention,
for example, by a user device.
[0037] As earlier noted, in some embodiments, at a communication network or a
voice-
enabled device, such as, without limitation, communication networks 104, 204
and
devices 102, 202, respectively, a model may be trained with various sentence
features. For
example, the model may be trained with the earlier-enumerated language
attributes. Once the
model has been trained, devices 102, 202 may utilize the model to generate
language
attributes for a given sequence of inputted utterances. FIG. 4 shows an
example table 400 of
an output that devices 102, 202 may generate by employing one or more speech
detection
techniques or algorithms upon a sequence of utterances, in accordance with
some disclosed
embodiments. In some aspects, the utterance (or sentence) structure features
shown in FIG. 4
may be used to train a model of various disclosed embodiments and methods.
[0038] Example types of algorithms that devices 102, 202 may employ include,
without
limitation, algorithms that determine whether each term in a query is a "WH"
term (e.g.,
based on text generated from the utterances), determine whether each term in
the query is an
article (e.g., "a" or "the"), determine a part-of-speech for each term of the
query, and
determine the syllables of each term in the query. In some examples, the "WH"
terms and
article detection may be performed by processing text strings that are
generated from the
utterances. Example parts of speech algorithms that devices 102, 202 may
employ, for
instance, include those that are provided by the Natural Language Toolkit
(NLTK), spaCy,
and/or other natural language processing providers. Some of such algorithms
train parts of
speech models using classifiers such as DecisionTree, vectorizers, and/or the
like. In one
example, syllables are extracted from utterances by using a raw audio signal
to detect multiple
audio features and voice activity. Praat/Praat-Parselmouth is one example of
an open source
tool kit that may be employed for such syllable extraction. In another
example, an Ancient
Soundex algorithm can extract syllables from utterances by using text
generated based on the
utterances. Metaphone, Double metaphone, and Metaphone-3 are example
algorithms that
may perform text-based syllable extraction.
CA 03143933 2021-12-16
WO 2021/216135 - 11 - PC
T/US2020/066932
[0039] Table 400 includes columns 404 with each column including a word of the
phrase
"What is the time, TIVO?" , for example, uttered by user 106 or user 206 of
FIGS. 1 and 2,
respectively. Table 400 further includes rows 402, with each row representing
a tag or a
training feature. For example, the first row is for the feature "WH," the
second row is for the
feature "articles," the third row is for the feature "POS" and the fourth row
is for the feature
"syllables." An acoustic model may be trained with a set of features that are
in part or in
whole different than the feature set of FIG. 4, or the model may be trained
with a feature set
that includes less than four or more than four features. In general, the
greater the number of
sentence features the model trains with, the greater the accuracy of sentence
prediction.
[0040] Table 400 entries are marked based on the feature corresponding to each
word of the
sentence "What is the time, TIVO?". For example, "What" corresponds to the
feature "WH"
but the word "is" or the word "the" or "time" do not. Accordingly, a checkmark
is placed in
the entry of table 400 at the first row and first column. Similarly, the word
"the" is an article
and marked accordingly in the second row, third column of Table 400 and so on.
In this
respect, an acoustic model is trained to predict the words of a sentence and
therefore the entire
sentence. In a practical example, the model may be used to predict the words
of a sentence at
step 312 of process 300 (FIG. 3) and step 510 of FIG. 5.
[0041] FIG. 5 shows an illustrative flowchart of a speech recognition process,
in accordance
with some embodiments of the disclosure. In FIG. 5, a process 500 may be
performed by a
voice-assisted device, such as devices 102 and 202 of FIGS. 1 and 2,
respectively, to process
incoming speech data packets. In some embodiments, the steps of process 500
may be
performed by control circuitry 704 of FIG. 7. In summary, process 500 presents
an example of
a method for detecting a spoken sentence in a speech recognition system as
disclosed herein.
Speech data is buffered based on an audio signal captured at a control
circuitry operating in an
active mode. The speech data is buffered irrespective of whether the speech
data comprises a
signature word. The buffered speech data is processed to detect the presence
of a sentence
comprising at least one command and a query for the computing device.
Processing the
buffered speech data includes detecting the signature word in the buffered
speech data, and, in
response to detecting the signature word in the speech data, initiating
detection of the sentence
in the buffered speech data.
[0042] More specifically and with reference to FIG. 5, at 502, process 500
starts and
continues to step 504 where packets of speech data, corresponding to a user-
spoken sentence,
are buffered based on an audio signal captured in an active mode, as earlier
described. The
packets are previously received, for example, at audio signal receiver 120 or
receiver 220 of
CA 03143933 2021-12-16
WO 2021/216135 - 12 - PCT/US2020/066932
devices 102 and 202, respectively. While in active mode, the received data
packets may be
recorded in buffer 116 or buffer 216 of devices 101 and 102, respectively.
Next, at step 506, the
buffered speech data packets are processed. The voice-assisted device, such as
may be
implemented by control circuitry 704 (FIG. 7), detects the signature word at
step 508, followed,
at step 510, by initiating detection of the sentence in the buffered speech
data, in response to
detecting the signature word at step 508, in step 510. Steps 508 and 510 are
part of the
processing that starts at step 508. Processing is performed while the device
remains in active
mode. In some embodiments, the device leaves the active mode in response to a
manual
configuration, such as in response to receiving a corresponding user device
signal. In some
embodiments, the device may leave an active mode if a signature word is not
found during a
predefined time period at step 508. In some embodiments, the device leaves the
active mode in
response to receiving speech data packets corresponding to an entire spoken
sentence.
[0043] FIG. 6 shows an illustrative flowchart of a speech recognition process,
in accordance
with some embodiments of the disclosure. In FIG. 6, a process 600 may be
performed by a
remotely located (relative to a communicatively coupled voice-assisted device)
processor, such
as processor 124 of FIG. 1 or processor 224 of FIG. 2. Process 600 begins at
602 and continues
to step 604 where an audio file with recorded packets of speech data
corresponding to at least
one spoken sentence is received. In FIG. 6, the audio file is presumed to
include N number of
packets, "N" representing an integer value. In some embodiments, the audio
file of step 604
may be received from device 102 or device 202. Next, at step 606, the
beginning and ending of
the sentence in the audio file of step 604 are identified. If, at step 608,
process 600 determines
that all N sentences of the audio file have been processed, process 600
continues to step 604 and
starts to process the next audio file after it is received as previously
described. If, at step 608,
process 600 determines not all sentences of the audio file have been
processed, process 600
proceeds to step 610. At step 610, the current sentence, the sentence
identified at step 606, is
processed and next, at step 612, the processing of the next sentence of the
audio file begins, and
the "current" sentence of the following steps in process 600, i.e., steps 604
through 610, is the
next sequential sentence in the audio file. In some embodiments, phrases of an
audio file need
not be sequentially processed. For example, phrase 3 may be processed before
phrase 2 in
FIG. 2. But in certain implementations using context speech recognition
techniques, the
accuracy of sentence prediction may improve if the sentences are sequentially
processed.
[0044] At step 610, the current sentence may be transmitted to a remote
automated speech
recognition (ASR) service for text transcription. In some embodiments, ASR
services may be
CA 03143933 2021-12-16
WO 2021/216135 - 13 - PCT/US2020/066932
performed on the audio file after all sentences of the file have been
processed. In process 600,
ASR services are presumed performed on a sentence basis rather than on an
audio file basis.
[0045] The order of steps of each of the processes 300, 500 and 600, as shown
in the
flowcharts of FIGS. 3, 5, and 6, respectively, may be suitably changed or
exchanged. One or
more steps, as may be suitable, can be added or deleted to each of the
processes 300, 500,
and 600.
[0046] A user may access, process, transmit and receive content, in addition
to other
features, for example to carry out the functions and implementations shown and
described
herein, with one or more user devices (i.e., user equipment). FIG. 7 shows
generalized
embodiments of an illustrative user device. In some embodiments, user device
700 may be
configured, in whole or in part, as a computing device. Although illustrated
as a mobile user
device (e.g., a smartphone), user device 700 may include any user electronic
device that
performs speech recognition operations as disclosed herein. In some
embodiments, user
device 700 may incorporate, in part or in whole, or be communicatively coupled
to, each of
devices 102 and 202 of FIGS. 1 and 2. In some embodiments, user device 700 may
include a
desktop computer, a tablet, a laptop, a remote server, any other suitable
device, or any
combination thereof, for speech detection and recognition processing, as
described above, or
accessing content, such as, without limitation, wearable devices with
projected image
reflection capability, such as a head-mounted display (HMD) (e.g., optical
head-mounted
display (OHMD)), electronic devices with computer vision features, such as
augmented
reality (AR), virtual reality (VR), extended reality (XR), or mixed reality
(MR), portable hub
computing packs, a television, a Smart TV, a set-top box, an integrated
receiver decoder
(IRD) for handling satellite television, a digital storage device, a digital
media receiver
(DMR), a digital media adapter (DMA), a streaming media device, a DVD player,
a DVD
recorder, a connected DVD, a local media server, a BLU-RAY player, a BLU-RAY
recorder,
a personal computer (PC), a laptop computer, a tablet computer, a WebTV box, a
personal
computer television (PC/TV), a PC media server, a PC media center, a handheld
computer, a
stationary telephone, a personal digital assistant (PDA), a mobile telephone,
a portable video
player, a portable music player, a portable gaming machine, a smartphone, or
any other
television equipment, computing equipment, or wireless device, and/or
combination of the
same. In some embodiments, the user device may have a front-facing screen and
a rear-facing
screen, multiple front screens, or multiple angled screens. In some
embodiments, the user
device may have a front-facing camera and/or a rear-facing camera. On these
user devices,
users may be able to navigate among and locate the same content available
through a
CA 03143933 2021-12-16
WO 2021/216135 - 14 - PCT/US2020/066932
television. Consequently, a user interface in accordance with the present
disclosure may be
available on these devices, as well. The user interface may be for content
available only
through a television, for content available only through one or more of other
types of user
devices, or for content available both through a television and one or more of
the other types
of user devices. The user interfaces described herein may be provided as
online applications
(i.e., provided on a website), or as stand-alone applications or clients on
user equipment
devices. Various devices and platforms that may implement the present
disclosure are
described in more detail below.
[0047] In some embodiments, display 712 may include a touchscreen, a
television display
or a computer display. In a practical example, display 712 may display
detected phrases from
user utterances, as processed by devices 102 and 202 or at communication
networks 104
and 204. Alternatively, or additionally, display 712 may show a respective
user the terms of a
user privacy agreement, as previously discussed relative to FIGS. 1 and 2.
Display 712 may
optionally show text results received from an ASR service. In some
embodiments, the one or
more circuit boards illustrated include processing circuitry, control
circuitry, and storage (e.g.,
RAM, ROM, Hard Disk, Removable Disk, etc.). In some embodiments, the
processing
circuit, control circuitry, or a combination thereof, may implement one or
more of the
processes of FIGS. 3, 5, and 6. In some embodiments, the processing circuitry,
control
circuitry, or a combination thereof, may implement one or more functions or
components of
the devices of FIGS. 1 and 2, such as devices 102 and 202, and/or processors
124 and 224.
For example, each or a combination of activity detector 118 or 218 and
processor 124 or 224
of FIGS. 1 and 2 may be implemented by the processing circuitry, control
circuitry or a
combination of the processing circuitry and control circuitry.
[0048] In some embodiments, circuit boards include an input/output path. User
device 700
may receive content and data via input/output (hereinafter "I/0") path 702.
I/0 path 702 may
provide content and data to control circuitry 704, which includes processing
circuitry 706 and
storage 708. Control circuitry 704 may be used to send and receive commands,
requests, and
other suitable data using I/0 path 702. I/0 path 702 may connect control
circuitry 704 (and
specifically processing circuitry 706) to one or more communications paths
(described
below). I/0 functions may be provided by one or more of these communications
paths but are
shown as a single path in FIG. 7 to avoid overcomplicating the drawing.
[0049] Control circuitry 704 may be based on any suitable processing circuitry
such as
processing circuitry 706. As referred to herein, processing circuitry should
be understood to
mean circuitry based on one or more microprocessors, microcontrollers, digital
signal
CA 03143933 2021-12-16
WO 2021/216135 - 15 - PCT/US2020/066932
processors, programmable logic devices, field-programmable gate arrays
(FPGAs),
application-specific integrated circuits (ASICs), etc., and may include a
multi-core processor
(e.g., dual-core, quad-core, hexa-core, or any suitable number of cores) or
supercomputer. In
some embodiments, processing circuitry is distributed across multiple separate
processors or
processing units, for example, multiple of the same type of processing units
(e.g., two Intel
Core i7 processors) or multiple different processors (e.g., an Intel Core i5
processor and an
Intel Core i7 processor). In some embodiments, control circuitry 704 executes
instructions
for an application stored in memory (e.g., storage 708). Specifically, control
circuitry 704
may be instructed by the application to perform the functions discussed above
and below. For
example, the application may provide instructions to control circuitry 704 to
perform speech
detection and recognition processes as described herein. In some
implementations, any action
performed by control circuitry 704 may be based on instructions received from
the
application.
[0050] In some client/server-based embodiments, control circuitry 704 includes
communications circuitry suitable for communicating with an application server
or other
networks or servers. The instructions for carrying out the above-mentioned
functionality may
be stored on the application server. Communications circuitry may include a
wired or wireless
modem or an ethernet card for communications with other equipment, or any
other suitable
communications circuitry. Such communications may involve the Internet or any
other
suitable communications networks or paths. In addition, communications
circuitry may
include circuitry that enables peer-to-peer communication of user equipment
devices, or
communication of user equipment devices in locations remote from each other
(described in
more detail below).
[0051] Memory may be an electronic storage device provided as storage 708 that
is part of
control circuitry 704. As referred to herein, the phrase "electronic storage
device" or "storage
device" or "memory" should be understood to mean any device for storing
electronic data,
computer software, or firmware, such as random-access memory, read-only
memory, hard
drives, optical drives, solid state devices, quantum storage devices, gaming
consoles, gaming
media, or any other suitable fixed or removable storage devices, and/or any
combination of
the same. Storage 708 may be used to store various types of content described
herein as well
as media guidance data described above. Nonvolatile memory may also be used
(e.g., to
launch a boot-up routine and other instructions). Cloud-based storage, for
example, may be
used to supplement storage 708 or instead of storage 708. In some embodiments,
storage 708
may incorporate, in part or in whole, buffer 116 and buffer 216 of FIGS. 1 and
2, respectively.
CA 03143933 2021-12-16
WO 2021/216135 - 16 - PCT/US2020/066932
[0052] In some embodiments, display 712 is caused by generation of a display
by devices 102
and 202 of FIGS. 1 and 2, respectively, or user devices coupled to devices 102
and 202. A
user may send instructions to control circuitry 704 using user input interface
710. User input
interface 710, display 712, or both may include a touchscreen configured to
provide a display
and receive haptic input. For example, the touchscreen may be configured to
receive haptic
input from a finger, a stylus, or both. In some embodiments, equipment device
700 may
include a front-facing screen and a rear-facing screen, multiple front
screens, or multiple
angled screens. In some embodiments, user input interface 710 includes a
remote-control
device having one or more microphones, buttons, keypads, any other components
configured
to receive user input or combinations thereof For example, user input
interface 710 may
include a handheld remote-control device having an alphanumeric keypad and
option buttons.
[0053] Audio equipment 714 may be provided as integrated with other elements
of user
device 700 or may be stand-alone units. The audio component of videos and
other content
displayed on display 712 may be played through speakers of audio equipment
714. In some
embodiments, the audio may be distributed to a receiver (not shown), which
processes and
outputs the audio via speakers of audio equipment 714. In some embodiments,
for example,
control circuitry 704 is configured to provide audio cues to a user, or other
audio feedback to
a user, using speakers of audio equipment 714. Audio equipment 714 may include
a
microphone configured to receive audio input such as voice commands or speech.
For
example, a user may speak letters or words that are received by the microphone
and converted
to text by control circuitry 704. In a further example, a user may voice
commands that are
received by the microphone and recognized by control circuitry 704.
[0054] An application may be implemented using any suitable architecture. For
example, a
stand-alone application may be wholly implemented on user device 700. In some
such
embodiments, instructions for the application are stored locally (e.g., in
storage 708), and data
for use by the application is downloaded on a periodic basis (e.g., from an
out-of-band feed,
from an Internet resource, or using another suitable approach). Control
circuitry 704 may
retrieve instructions of the application from storage 708 and process the
instructions to
generate any of the displays discussed herein. Based on the processed
instructions, control
circuitry 704 may determine what action to perform when input is received from
input
interface 710. For example, movement of a cursor on a display up/down may be
indicated by
the processed instructions when input interface 710 indicates that an up/down
button was
selected. An application and/or any instructions for performing any of the
embodiments
discussed herein may be encoded on computer-readable media. Computer-readable
media
CA 03143933 2021-12-16
WO 2021/216135 - 17 - PCT/US2020/066932
includes any media capable of storing data. The computer-readable media may be
transitory,
including, but not limited to, propagating electrical or electromagnetic
signals, or it may be
non-transitory including, but not limited to, volatile and non-volatile
computer memory or
storage devices such as a hard disk, floppy disk, USB drive, DVD, CD, media
cards, register
memory, processor caches, Random Access Memory (RAM), etc.
[0055] In some embodiments, the application is a client/server-based
application. Data for
use by a thick or thin client implemented on user device 700 is retrieved on
demand by
issuing requests to a server remote from user device 700. For example, the
remote server may
store the instructions for the application in a storage device. The remote
server may process
the stored instructions using circuitry (e.g., control circuitry 704) and
generate the displays
discussed above and below. The client device may receive the displays
generated by the
remote server and may display the content of the displays locally on user
device 700. This
way, the processing of the instructions is performed remotely by the server
while the resulting
displays (e.g., that may include text, a keyboard, or other visuals) are
provided locally on user
device 700. User device 700 may receive inputs from the user via input
interface 610 and
transmit those inputs to the remote server for processing and generating the
corresponding
displays. For example, user device 700 may transmit a communication to the
remote server
indicating that an up/down button was selected via input interface 710. The
remote server
may process instructions in accordance with that input and generate a display
of the
application corresponding to the input (e.g., a display that moves a cursor
up/down). The
generated display is then transmitted to user device 700 for presentation to
the user.
[0056] FIG. 8 is a block diagram of illustrative system 800 for transmitting
messages, in
accordance with some embodiments of the present disclosure. In system 800,
there may be
more than one of each type of user device, but only one of each is shown in
FIG. 8 to avoid
overcomplicating the drawing. In addition, each user may utilize more than one
type of user
device and more than one of each type of user device.
[0057] User device 820, illustrated as a wireless-enabled device, may be
coupled to
communication network 802 (e.g., the Internet). For example, user device 820
is coupled to
communication network 802 via communications path 822 to access point 824 and
wired
connection 826. User device 820 may also include wired connections to a LAN,
or any other
suitable communications link to network 802. Communication network 802 may be
one or
more networks including the Internet, a mobile phone network, mobile voice or
data network
(e.g., a WIFI, WiMAX, GSM, UTMS, CDMA, TDMA, 3G, 4G, 4G, 5G, Li-Fi, LTE
network), cable network, public switched telephone network, or other types of
communication
CA 03143933 2021-12-16
WO 2021/216135 - 18 - PCT/US2020/066932
network or combinations of communication networks. Path 812 may include one or
more
communications paths, such as a satellite path, a fiber-optic path, a cable
path, a path that
supports Internet communications, a free-space connection (e.g., for broadcast
or other
wireless signals), or any other suitable wired or wireless communications path
or combination
.. of such paths.
[0058] System 800 includes network entity 804 (e.g., a server or other
suitable computing
device) coupled to communication network 802 via communications path 812.
Communications with network entity 804 may be exchanged over one or more
communications paths but are shown as a single path in FIG. 8 to avoid
overcomplicating the
.. drawing. Network entity 804 is configured to access database 806 or
applications 808 (e.g.,
an applications database or host server) via communications links 814 and 816,
respectively.
Although shown as a separate device, network entity 804 may include database
806 and
applications 808 (e.g., stored in memory). In addition, there may be more than
one of each of
database 806 and applications 808, but only one of each is shown in FIG. 8 to
avoid
overcomplicating the drawing. In some embodiments, database 806 and
applications 808 may
be integrated as one source device (e.g., that may be, but need not be,
network entity 804).
[0059] Database 806 may include one or more types of stored information,
including, for
example, relationship information, a relationship entity database, recipient
information,
historical communications records, user preferences, user profile information,
a template
.. database, any other suitable information, or any combination thereof.
Applications 808 may
include an applications-hosting database or server, plug-ins, a software
developers kit (SDK),
an applications programming interface (API), or other software tools
configured to provide
software (e.g., as download to a user device); run software remotely (e.g.,
hosting applications
accessed by user devices); or otherwise provide applications support to
applications of user
device 820. In some embodiments, information from network entity 804, database
806,
applications 808, or a combination thereof may be provided to a user device
using a
client/server approach. For example, user device 820 may pull information from
a server, or a
server may push information to user device 820. In some embodiments, an
application client
residing on user device 820 may initiate sessions with database 806,
applications 808,
.. network entity 804, or a combination thereof to obtain information when
needed (e.g., when
data is out-of-date or when a user device receives a request from the user to
receive data). In
some embodiments, information may include user information. For example, the
user
information may include current and/or historical user activity information
(e.g., what
communications the user engages in, what times of day the user sends/receives
messages,
CA 03143933 2021-12-16
WO 2021/216135 - 19 - PCT/US2020/066932
whether the user interacts with a social network, at what times the user
interacts with a social
network to post information, what types of content the user typically inserts
in messages,
stored contacts of the user, frequent contacts of the user, any other suitable
information, or
any combination thereof. In some embodiments, the user information may
identify patterns
of a given user for a period of more than one year.
[0060] In some embodiments, an application may include an application program
processor
implementing some of the processes and methods disclosed herein as a stand-
alone
application implemented on user device 820. For example, the application may
be
implemented as software or a set of executable instructions, which may be
stored in storage
(e.g., storage 708) of the user device (e.g., user device 700), and executed
by control circuitry
(e.g., control circuitry 704) of the user device (e.g., user device 700). In
some embodiments,
an application may include an automatic program retrieval application that is
implemented as
a client/server-based application where only a client application resides on
the user device,
and a server application resides on a remote server (e.g., network entity
804). For example,
an automatic program retrieval application may be implemented partially as a
client
application on user device 820 (e.g., by control circuitry 704 of user
equipment device 700)
and partially on a remote server as a server application running on control
circuitry of the
remote server (e.g., control circuitry of network entity 804). When executed
by control
circuitry of the remote server, the automatic program retrieval application
may instruct the
control circuitry to generate the displays and transmit the generated displays
to user
device 820. The server application may instruct the control circuitry of the
remote device to
transmit data for storage on user device 820. The client application may
instruct control
circuitry of the receiving user device to generate the application displays.
[0061] In some embodiments, the arrangement of system 800 is a cloud-based
arrangement.
The cloud provides access to services, such as information storage, messaging,
or social
networking services, among other examples, as well as access to any content
described above,
for user devices. Services can be provided in the cloud through cloud
computing service
providers, or through other providers of online services. For example, the
cloud-based
services can include a storage service, a sharing site, a social networking
site, or other
services via which user-sourced content is distributed for viewing by others
on connected
devices. These cloud-based services may allow a user device to store
information to the cloud
and to receive information from the cloud rather than storing information
locally and
accessing locally stored information. Cloud resources may be accessed by a
user device
using, for example, a web browser, a messaging application, a desktop
application, a mobile
CA 03143933 2021-12-16
WO 2021/216135 - 20 - PCT/US2020/066932
application, and/or any combination of the same access applications. The user
device may be
a cloud client that relies on cloud computing for application delivery, or the
user equipment
device may have some functionality without access to cloud resources. For
example, some
applications running on the user device may be cloud applications (e.g.,
applications delivered
as a service over the Internet), while other applications may be stored and
run on the user
device. In some embodiments, a user device may receive information from
multiple cloud
resources simultaneously.
[0062] The systems and processes discussed above are intended to be
illustrative and not
limiting. One skilled in the art would appreciate that the actions of the
processes discussed
herein may be omitted, modified, combined, and/or rearranged, and any
additional actions
may be performed without departing from the scope of the invention. More
generally, the
above disclosure is meant to be exemplary and not limiting. Only the claims
that follow are
meant to set bounds as to what the present disclosure includes. Furthermore,
it should be
noted that the features and limitations described in any one embodiment may be
applied to
any other embodiment herein, and flowcharts or examples relating to one
embodiment may be
combined with any other embodiment in a suitable manner, done in different
orders, or done
in parallel. In addition, the systems and methods described herein may be
performed in real
time. It should also be noted that the systems and/or methods described above
may be applied
to, or used in accordance with, other systems and/or methods.
CA 03143933 2021-12-16
WO 2021/216135 - 21 - PCT/US2020/066932
This specification discloses embodiments which include, but are not limited
to, the
following:
1. A method for processing speech in a speech recognition system, the method
comprising:
detecting a user event indicative of a user intention to interact with a
speech
recognition device;
in response to detecting the user event, enabling an active mode of the speech
recognition device to record speech data based on an audio signal captured at
the speech
recognition device irrespective of whether the speech data comprises a
signature word; and
while the active mode is enabled:
generating a recording of the speech data;
detecting the signature word in a portion of the speech data other than a
beginning portion of the speech data; and
in response to detecting the signature word, processing the recording of the
speech data to recognize a user-uttered phrase.
2. The method of item 1, wherein generating the recording is performed at the
speech
recognition device.
3. The method of item 1, wherein processing the recording of the speech data
is performed at
a server remote from the speech recognition device.
4. The method of item 1, wherein detecting the signature word is performed
based on an
acoustic model.
5. The method of item 4, wherein the acoustic model is selected from one of a
hidden Markov
model (HMM), a long short-term memory (LSTM) model, and a bidirectional LSTM.
6. The method of item 1, wherein detecting the signature word is based on
heuristics of audio
signatures of a demographic region.
7. The method of item 1, further comprising determining whether the speech
data corresponds
to human speech based on a spectral characteristic analysis of the audio
signal captured at the
speech recognition device.
CA 03143933 2021-12-16
WO 2021/216135 - 22 - PCT/US2020/066932
8. The method of item 7, further comprising determining whether the speech
data corresponds
to human speech based on a comparison of the audio signal captured at the
speech recognition
device and a list of black-listed audio signals.
9. The method of item 1, wherein detecting the user event comprises detecting
a user activity
suggestive of a user movement in closer proximity to the speech recognition
device.
10. The method of item 9, wherein detecting the user activity comprises
sensing the user
movement with a device selected from one or more of a motion detector device,
an infrared
.. recognition device, an ultraviolet-based detection device, and an image
capturing device.
11. A system for processing speech in a speech recognition system, the system
comprising:
a sensor configured to detect a user event indicative of a user intention to
interact with
a speech recognition device;
a memory; and
control circuitry communicatively coupled to the memory and the sensor and
configured to:
in response to detecting the user event, enable an active mode of the speech
recognition device to record, in the memory, speech data based on an audio
signal
captured at the speech recognition device irrespective of whether the speech
data
comprises a signature word; and
while the active mode is enabled:
generate a recording of the speech data;
detect the signature word in a portion of the speech data other than a
beginning portion of the speech data; and
in response to detecting the signature word, process the recording of the
speech data to recognize a user-uttered phrase.
12. The system of item 11, wherein the control circuitry is configured to
generate the
.. recording at the speech recognition device.
13. The system of item 11, wherein the control circuitry is configured to
process the recording
of the speech data by causing the recording to be processed at a server remote
from the speech
recognition device.
CA 03143933 2021-12-16
WO 2021/216135 - 23 - PCT/US2020/066932
14. The system of item 11, wherein the control circuitry is configured to
detect the signature
word based on an acoustic model.
15. The system of item 14, wherein the acoustic model is selected from one of
a hidden
Markov model (HMM), a long short-term memory (LSTM) model, and a bidirectional
LSTM.
16. The system of item 11, wherein the control circuitry is configured to
detect the signature
word based on heuristics of audio signatures of a demographic region.
17. The system of item 11, wherein the control circuitry is further configured
to determine
whether the speech data corresponds to human speech based on a spectral
characteristic
analysis of the audio signal captured at the speech recognition device.
18. The system of item 17, wherein the control circuitry is further configured
to determine
whether the speech data corresponds to human speech based on a comparison of
the audio
signal captured at the speech recognition device and a list of black-listed
audio signals.
19. The system of item 11, wherein the control circuitry is configured to
detect the user event
by detecting a user activity suggestive of a user movement in closer proximity
to the speech
recognition device.
20. The system of item 19, wherein the control circuitry is configured to
detect the user
activity by sensing the user movement with a device selected from one or more
of a motion
detector device, an infrared recognition device, an ultraviolet-based
detection device, and an
image capturing device.
21. A non-transitory computer-readable medium having instructions encoded
thereon that
when executed by control circuitry cause the control circuitry to:
detect a user event indicative of a user intention to interact with a speech
recognition
device;
in response to detecting the user event, enable an active mode of the speech
recognition device to record speech data based on an audio signal captured at
the speech
recognition device irrespective of whether the speech data comprises a
signature word; and
while the active mode is enabled:
generate a recording of the speech data;
CA 03143933 2021-12-16
WO 2021/216135 - 24 - PCT/US2020/066932
detect the signature word in a portion of the speech data other than a
beginning
portion of the speech data; and
in response to detecting the signature word, process the recording of the
speech
data to recognize a user-uttered phrase.
22. The non-transitory computer-readable medium of item 21, wherein generating
the
recording is performed at the speech recognition device.
23. The non-transitory computer-readable medium of item 21, wherein processing
the
recording of the speech data is performed at a server remote from the speech
recognition
device.
24. The non-transitory computer-readable medium of item 21, wherein detecting
the signature
word is performed based on an acoustic model.
25. The non-transitory computer-readable medium of item 24, wherein the
acoustic model is
selected from one of a hidden Markov model (HMM), a long short-term memory
(LSTM)
model, and a bidirectional LSTM.
26. The non-transitory computer-readable medium of item 21, wherein detecting
the signature
word is based on heuristics of audio signatures of a demographic region.
27. The non-transitory computer-readable medium of item 21, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
determine whether the speech data corresponds to human speech based on a
spectral
characteristic analysis of the audio signal captured at the speech recognition
device.
28. The non-transitory computer-readable medium of item 27, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
determine whether the speech data corresponds to human speech based on a
comparison of
the audio signal captured at the speech recognition device and a list of black-
listed audio
signals.
29. The non-transitory computer-readable medium of item 21, wherein detecting
the user
event comprises detecting a user activity suggestive of a user movement in
closer proximity to
the speech recognition device.
CA 03143933 2021-12-16
WO 2021/216135 - 25 -
PCT/US2020/066932
30. The non-transitory computer-readable medium of item 29, wherein detecting
the user
activity comprises sensing the user movement with a device selected from one
or more of a
motion detector device, an infrared recognition device, an ultraviolet-based
detection device,
and an image capturing device.
31. A system for processing speech in a speech recognition system, the system
comprising:
means for detecting a user event indicative of a user intention to interact
with a speech
recognition device;
means for, in response to detecting the user event, enabling an active mode of
the
speech recognition device to record speech data based on an audio signal
captured at the
speech recognition device irrespective of whether the speech data comprises a
signature word;
and
means for, while the active mode is enabled:
generating a recording of the speech data;
detecting the signature word in a portion of the speech data other than a
beginning portion of the speech data; and
in response to detecting the signature word, processing the recording of the
speech data to recognize a user-uttered phrase.
32. The system of item 31, wherein the means for generating the recording is
located at the
speech recognition device.
33. The system of item 31 or 32, wherein the means for processing the
recording of the speech
data is located at a server remote from the speech recognition device.
34. The system of item 31, 32 or 33, wherein the means for detecting the
signature word is
configured to perform the detecting based on an acoustic model.
35. The system of item 34, wherein the acoustic model is selected from one of
a hidden
Markov model, a long short-term memory model, and a bidirectional long short-
term memory
model.
CA 03143933 2021-12-16
WO 2021/216135 - 26 - PCT/US2020/066932
36. The system of any of items 31-35, wherein means for detecting the
signature word is
configured to detect the signature based on heuristics of audio signatures of
a demographic
region.
37. The system of any of items 31-36, further comprising means for determining
whether the
speech data corresponds to human speech based on a spectral characteristic
analysis of the
audio signal captured at the speech recognition device.
38. The system of item 37, further comprising means for determining whether
the speech data
.. corresponds to human speech based on a comparison of the audio signal
captured at the
speech recognition device and a list of black-listed audio signals.
39. The system of any of items 31-38, wherein the means for detecting the user
event
comprises means for detecting a user activity suggestive of a user movement in
closer
proximity to the speech recognition device.
40. The system of item 39, wherein the means for detecting the user activity
comprises means
for sensing the user movement with a device selected from one or more of a
motion detector
device, an infrared recognition device, an ultraviolet-based detection device,
and an image
capturing device.
41. A method of processing speech in a speech recognition system, the method
comprising:
detecting a user event indicative of a user intention to interact with a
speech
recognition device;
in response to detecting the user event, enabling an active mode of the speech
recognition device to record speech data based on an audio signal captured at
the speech
recognition device irrespective of whether the speech data comprises a
signature word; and
while the active mode is enabled:
generating a recording of the speech data;
detecting the signature word in a portion of the speech data other than a
beginning portion of the speech data; and
in response to detecting the signature word, processing the recording of the
speech data to recognize a user-uttered phrase.
CA 03143933 2021-12-16
WO 2021/216135 - 27 - PCT/US2020/066932
42. The method of item 41, wherein generating the recording is performed at
the speech
recognition device.
43. The method of item 41 or 42, wherein processing the recording of the
speech data is
performed at a server remote from the speech recognition device.
44. The method of any of items 41 to 43, wherein detecting the signature word
is performed
based on an acoustic model.
45. The method of item 44, wherein the acoustic model is selected from one of
a hidden
Markov model, a long short-term memory model, and a bidirectional long short-
term memory
model.
46. The method of any of items 41 to 45, wherein detecting the signature word
is based on
heuristics of audio signatures of a demographic region.
47. The method of any of items 41 to 46, further comprising determining
whether the speech
data corresponds to human speech based on a spectral characteristic analysis
of the audio
signal captured at the speech recognition device.
48. The method of item 47, further comprising determining whether the speech
data
corresponds to human speech based on a comparison of the audio signal captured
at the
speech recognition device and a list of black-listed audio signals.
49. The method of any of items 41 to 48, wherein detecting the user event
comprises detecting
a user activity suggestive of a user movement in closer proximity to the
speech recognition
device.
50. The method of item 49, wherein detecting the user activity comprises
sensing the user
movement with a device selected from one or more of a motion detector device,
an infrared
recognition device, an ultraviolet-based detection device, and an image
capturing device.
51. A computer program comprising computer-readable instructions that, when
executed by
one or more processors, causes the one or more processors to perform the
method of any of
items 41-50.
CA 03143933 2021-12-16
WO 2021/216135 - 28 - PCT/US2020/066932
52. A method for detecting a sentence including at least one of a command and
a query in a
speech recognition system, the method comprising:
buffering speech data based on an audio signal captured at a computing device
operating in an active mode, wherein the speech data is buffered irrespective
of whether the
speech data comprises a signature word; and
processing the buffered speech data to detect a presence of the sentence
comprising at
least one of the command and the query for the computing device, wherein
processing the
buffered speech data comprises:
detecting the signature word in the buffered speech data, and
in response to detecting the signature word in the speech data, initiating
detection of the sentence in the buffered speech data.
53. The method of item 52, further comprising detecting the signature word
based on a
sequence validating technique.
54. The method of item 52, further comprising detecting the signature word
based on a model
trained to distinguish between user commands and user assertions.
55. The method of item 52, further comprising detecting the signature word by
detecting
silent durations occurring before and after, respectively, the sentence in the
speech data.
56. The method of item 55, wherein detecting the silent durations is based on
speech
amplitude heuristics of the speech data.
57. The method of item 52, wherein detecting the signature word is performed
at the
computing device.
58. The method of item 52, wherein detecting the signature word is performed
at a server
remote from the computing device.
59. The method of item 52, further comprising transmitting the speech data to
a speech
recognition processor for performing automated speech recognition (ASR) on the
speech data.
CA 03143933 2021-12-16
WO 2021/216135 - 29 - PCT/US2020/066932
60. The method of item 52, further comprising detecting the signature word by
detecting a
beginning portion of the sentence and an end portion of the sentence based on
a trained model
selected from one of a hidden Markov model (HMM), a long short-term memory
(LSTM)
model, and a bidirectional LSTM.
61. The method of item 52, wherein detecting the signature word is based on
heuristics of
audio signatures of a demographic region.
62. The method of item 52, wherein the control circuitry operates in the
active mode only in
response to receiving a user consent.
63. The method of item 52, further comprising enabling the active mode in
response to a
detection of a user activity that suggests a user intention to interact with
the control circuitry.
64. A system for detecting a sentence including at least one of a command and
a query in a
speech recognition system, the system comprising:
a memory; and
control circuitry communicatively coupled to the memory and configured to:
buffer in the memory speech data based on an audio signal captured at a
computing device operating in an active mode, wherein the speech data is
buffered
irrespective of whether the speech data comprises a signature word; and
process the buffered speech data to detect a presence of the sentence
comprising at least one of the command and the query for the computing device;
wherein the control circuitry is configured to, in processing the buffered
speech data:
detect the signature word in the buffered speech data, and
in response to detecting the signature word in the speech data, initiate
detection of the sentence in the buffered speech data.
65. The system of item 64, wherein the control circuitry is further configured
to detect the
signature word based on a sequence validating technique.
66. The system of item 64, wherein the control circuitry is further configured
to detect the
signature word based on a model trained to distinguish between user commands
and user
assertions.
CA 03143933 2021-12-16
WO 2021/216135 - 30 - PCT/US2020/066932
67. The system of item 64, wherein the control circuitry is further configured
to detect the
signature word by detecting silent durations occurring before and after,
respectively, the
sentence in the speech data.
68. The system of item 67, wherein the control circuitry is further configured
to detect the
silent durations based on speech amplitude heuristics of the speech data.
69. The system of item 64, wherein the memory is local to the computing
device.
70. The system of item 64, wherein the control circuitry is further configured
to detect the
signature word at the computing device.
71. The system of item 64, wherein the control circuitry is configured to
transmit the buffered
data packets to a remotely located server to detect the signature word.
72. The system of item 64, wherein the control circuitry is further configured
to transmit the
speech data to a speech recognition processor for performing automated speech
recognition
(ASR) on the speech data.
73. The system of item 64, wherein the control circuitry is further configured
to detect the
signature word by detecting a beginning portion of the sentence and an end
portion of the
sentence based on a trained model selected from one of a hidden Markov model
(HMM), a
long short-term memory (LSTM) model, and a bidirectional LSTM.
74. The system of item 64, wherein detection of the signature word is based on
heuristics of
audio signatures of a demographic region.
75. The system of item 64, wherein the control circuitry is configured to
operate in the active
mode only in response to receiving a user consent.
76. The system of item 64, wherein the control circuitry is further configured
to enable the
active mode in response to a detection of a user activity that suggests a user
intention to
interact with the control circuitry.
CA 03143933 2021-12-16
WO 2021/216135 - 31 - PCT/US2020/066932
77. A non-transitory computer-readable medium having instructions encoded
thereon that
when executed by control circuitry cause the control circuitry to:
buffer speech data based on an audio signal captured at a computing device
operating
in an active mode, wherein the speech data is buffered irrespective of whether
the speech data
comprises a signature word; and
process the buffered speech data to detect a presence of the sentence
comprising at
least one of the command and the query for the computing device, wherein
processing the
buffered speech data comprises:
detecting the signature word in the buffered speech data, and
in response to detecting the signature word in the speech data, initiating
detection of the sentence in the buffered speech data.
78. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the signature word based on a sequence validating technique.
79. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the signature word based on a model trained to distinguish between user
commands and
user assertions.
80. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the signature word by detecting silent durations occurring before and
after,
respectively, the sentence in the speech data.
81. The non-transitory computer-readable medium of item 80, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the silent durations based on speech amplitude heuristics of the speech
data.
82. The non-transitory computer-readable medium of item 77, wherein detecting
the signature
word is performed at the computing device.
CA 03143933 2021-12-16
WO 2021/216135 - 32 - PCT/US2020/066932
83. The non-transitory computer-readable medium of item 77, wherein detecting
the signature
word is performed at a server remotely located from the computing device.
84. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
transmit the speech data to a speech recognition processor for performing
automated speech
recognition (ASR) on the speech data.
85. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the signature word by detecting a beginning portion of the sentence and
an end portion
of the sentence based on a trained model selected from one of a hidden Markov
model
(HMM), a long short-term memory (LSTM) model, and a bidirectional LSTM.
86. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
detect the signature word based on heuristics of audio signatures of a
demographic region.
87. The non-transitory computer-readable medium of item 77, wherein the
control circuitry
operates in the active mode only in response to receiving a user consent.
88. The non-transitory computer-readable medium of item 77, further having
instructions
encoded thereon that when executed by the control circuitry cause the control
circuitry to
enable the active mode in response to a detection of a user activity that
suggests a user
intention to interact with the control circuitry.
89. A system for detecting a sentence including at least one of a command and
a query in a
speech recognition system, the system comprising:
means for buffering speech data based on an audio signal captured at a
computing
device operating in an active mode, wherein the speech data is buffered
irrespective of
whether the speech data comprises a signature word; and
means for processing the buffered speech data to detect a presence of the
sentence
comprising at least one of the command and the query for the computing device,
wherein the
means for processing the buffered speech data comprises:
CA 03143933 2021-12-16
WO 2021/216135 - 33 - PCT/US2020/066932
means for detecting the signature word in the buffered speech data, and
means for, in response to detecting the signature word in the speech data,
initiating detection of the sentence in the buffered speech data.
90. The system of item 89, further comprising means for detecting the
signature word based
on a sequence validating technique.
91. The system of item 89 or 90, further comprising means for detecting the
signature word
based on a model trained to distinguish between user commands and user
assertions.
92. The system of item 89, 90 or 91, further comprising means for detecting
the signature
word by detecting silent durations occurring before and after, respectively,
the sentence in the
speech data.
93. The system of item 92, wherein the means for detecting the silent
durations comprises
means for detecting the silent durations based on speech amplitude heuristics
of the speech
data.
94. The system of any of items 89 to 93, wherein the means for detecting the
signature word
is located at the computing device.
95. The system of any of items 89 to 94, wherein the means for detecting the
signature word
is located at a server remote from the computing device.
96. The system of any of items 89 to 95, further comprising means for
transmitting the speech
data to a speech recognition processor for performing automated speech
recognition on the
speech data.
97. The system of any of items 89 to 96, further comprising means for
detecting the signature
word by detecting a beginning portion of the sentence and an end portion of
the sentence
based on a trained model selected from one of a hidden Markov model, a long
short-term
memory model, and a bidirectional long short-term memory model.
CA 03143933 2021-12-16
WO 2021/216135 - 34 - PCT/US2020/066932
98. The system of any of items 89 to 97, wherein the means for detecting the
signature word
comprises means for detecting the signature word based on heuristics of audio
signatures of a
demographic region.
99. The system of any of items 89 to 98, wherein the control circuitry
operates in the active
mode only in response to receiving a user consent.
100. The system of any of items 89 to 99, further comprising means for
enabling the active
mode in response to a detection of a user activity that suggests a user
intention to interact with
the control circuitry.
101. A method for detecting a sentence including at least one of a command and
a query in a
speech recognition system, the method comprising:
buffering speech data based on an audio signal captured at a computing device
operating in an active mode, wherein the speech data is buffered irrespective
of whether the
speech data comprises a signature word; and
processing the buffered speech data to detect a presence of the sentence
comprising at
least one of the command and the query for the computing device, wherein
processing the
buffered speech data comprises:
detecting the signature word in the buffered speech data, and
in response to detecting the signature word in the speech data, initiating
detection of the sentence in the buffered speech data.
102. The method of item 101, further comprising detecting the signature word
based on a
sequence validating technique.
103. The method of item 101 or 102, further comprising detecting the signature
word based
on a model trained to distinguish between user commands and user assertions.
104. The method of item 101, 102 or 103, further comprising detecting the
signature word by
detecting silent durations occurring before and after, respectively, the
sentence in the speech
data.
CA 03143933 2021-12-16
WO 2021/216135 - 35 - PCT/US2020/066932
105. The method of item 104, wherein detecting the silent durations is based
on speech
amplitude heuristics of the speech data.
106. The method of any of items 101 to 105, wherein detecting the signature
word is
performed at the computing device.
107. The method of any of items 101 to 105, wherein detecting the signature
word is
performed at a server remote from the computing device.
108. The method of any of items 101 to 107, further comprising transmitting
the speech data
to a speech recognition processor for performing automated speech recognition
on the speech
data.
109. The method of any one of items 101 to 108, further comprising detecting
the signature
word by detecting a beginning portion of the sentence and an end portion of
the sentence
based on a trained model selected from one of a hidden Markov model, a long
short-term
memory model, and a bidirectional long short-term memory model.
110. The method of any one of items 101-109, wherein detecting the signature
word is based
on heuristics of audio signatures of a demographic region.
111. The method of any one of items 101-110, wherein the control circuitry
operates in the
active mode only in response to receiving a user consent.
112. The method of any one of items 101-110, further comprising enabling the
active mode in
response to a detection of a user activity that suggests a user intention to
interact with the
control circuitry.
113. A computer program comprising computer readable instructions that, when
executed
by one or more processors, causes the one or more processors to perform the
method of any of
items 101-112.