Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
VOICE QUALITY INSPECTION METHOD AND DEVICE, COMPUTER
EQUIPMENT AND STORAGE MEDIUM
Technical Field
[0001] The present disclosure relates to information processing technical
field, particularly to
a voice quality inspection method, apparatus, computer device and storage
medium.
Background
[0002] With development of information technology, customers usually
communicate with
customer service to conduct operations such as business consulting and
complaining through
voice conversation. In order to improve voice service quality and customer
satisfaction,
enterprises need to evaluate and assess the performance of voice customer
service. At present,
the voice conversation records generated between customer service and customer
within a
period of time are sorted out and quality inspected mainly by manual means.
However, with
business development, the manual method has defects such as massive operation
steps, low
execution efficiency, poor quality inspection, high labor costs and high
subjectivity. Problems
such as low accuracy rate of voice quality inspection result and unreasonable
result of voice
customer service performance evaluation occurred.
Invention Content
[0003] Based on this, it is necessary to provide a voice quality inspection
method, apparatus,
computer device and storage medium, the method improves the performance of
manual voice
quality inspection.
[0004] On one hand, a voice quality inspection method comprises:
[0005] Converting voice data into text data, performing sentence segmentation
processing on
the text data, obtaining text fragments;
[0006] Configuring keyword type according to task type, retrieving keyword
corresponding
to the keyword type, and updating keyword library;
[0007] Selecting keyword text from the keyword library, comparing with the
text fragments,
obtaining matching information of keyword text;
[0008] According to comparison between the matching information of the keyword
text and
1
Date Regue/Date Received 2023-01-17
the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0009] According to comparison between the matching coefficient of the keyword
type and a
preset matching threshold, obtaining quality inspection result of the task
type.
[0010] In an embodiment, configuring keyword type according to task type,
retrieving
keyword corresponding to the keyword type, and updating keyword library, the
steps of
updating keyword library comprises:
[0011] Configuring keyword type according to task type, wherein the keyword
type includes:
standard keyword, forbidden keyword, emotion keyword;
[0012] According to meaning and application scenario of each keyword type,
obtaining
corresponding keyword;
[0013] According to the keyword, updating keyword database.
[0014] In an embodiment, selecting keyword text from the keyword library,
comparing with
the text fragments, obtaining matching information of keyword text, the steps
comprises:
[0015] According to the keyword library, selecting at least one keyword text
for each keyword
type to form a single configuration of each keyword type;
[0016] Perfoiming content matching and traversal from first clause of the text
fragment with
the single configuration, if the first clause matches with keyword text in the
single
configuration, recording serial number of the clause in the text segment, and
obtaining
matching position of the keyword text;
[0017] Recording keyword text quantity, wherein the keyword text forms a
matching
relationship between the clause and the single configuration, if the same
keyword text in the
single configuration repeats many times in the same clause, the keyword text
quantity is only
counted once, according to the keyword text quantity, obtaining matching times
of the keyword
text;
[0018] According to the matching position of the keyword text and the matching
times of the
keyword text, obtaining matching information of the keyword text.
[0019] In an embodiment, according to comparison between the matching
information of the
keyword text and the selected keyword text quantity, obtaining matching
coefficient of the
keyword type, comprising:
2
Date Regue/Date Received 2023-01-17
[0020] According to the matching information of keyword text, matching times
is sorted,
according to the maximum matching times, obtaining the number of maximum
matching times;
[0021] Dividing the number of maximum matching times by the selected keyword
text
quantity, obtaining the matching coefficient of a single configuration;
[0022] Setting sampling weight for each single configuration, obtaining
sampling coefficient
of single configuration through the sampling weight and the matching
coefficient of single
configuration, the mathematical expression of single configuration sampling
coefficient sp is:
[0023] sp = w * p
[0024] Wherein, sp is sampling coefficient of single configuration, w is
sampling weight of
single configuration, and p is matching coefficient of single configuration;
[0025] Comparing the single configuration sampling coefficient with the preset
matching
threshold of single configuration, obtaining matching coefficient of keyword
type, the
mathematical expression of matching coefficient S of keyword type is:
[0026] S = d (max (0, sp¨
d(sp¨t)
[0027] Wherein, S is matching coefficient of keyword type, sp is sampling
coefficient of
single configuration, t is matching threshold of single configuration, max()
means maximum
value, d() is differential operator.
[0028] In an embodiment, according to comparison between the matching
coefficient of the
keyword type and a preset matching threshold, obtaining quality inspection
result of the task
type, comprising:
[0029] According to comparison between matching coefficient of each keyword
type and
each preset matching threshold, obtaining quality inspection result.
[0030] In an embodiment, according to comparison between the matching
coefficient of the
keyword type and a preset matching threshold, obtaining quality inspection
result of the task
type, also comprising:
[0031] Setting sampling weight to each keyword type, according to the matching
coefficient
of sampling weight and the keyword type, obtaining sampling coefficient of the
keyword type.
[0032] Comparing sampling coefficient with a preset matching threshold,
obtaining quality
3
Date Regue/Date Received 2023-01-17
inspection result.
[0033] In an embodiment, converting voice data into text data, performing
sentence
segmentation processing on the text data, obtaining text fragments,
comprising:
[0034] Separating muted content and voice content in voice data, obtaining
time separation
tag;
[0035] Segmenting text data according to the time separation tag, adding
punctuation marks
at the end of sentence according to color vocabulary at the end of sentence;
[0036] Querying word count of text without punctuation marks in the text data,
if the text
without punctuation marks exceeds a preset word count threshold, adding
punctuation mark.
[0037] In another aspect, a voice quality inspection apparatus is provided,
the apparatus
comprises:
[0038] A voice converting module configured to convert voice data into text
data, perform
sentence segmentation processing on the text data, obtain text fragments;
[0039] A task configuration module configured to configure keyword type
according to task
type, retrieve keyword corresponding to the keyword type, and update keyword
library;
[0040] A matching information obtaining module configured to select keyword
text from the
keyword library, compare with the text fragments, and obtain matching
information of keyword
text;
[0041] A matching coefficient obtaining module configured to obtain matching
coefficient
of the keyword type according to comparison between the matching information
of the
keyword text and the selected keyword text quantity;
[0042] A quality inspection result obtaining module configured to obtain
quality inspection
result of the task type according to comparison between the matching
coefficient of the
keyword type and a preset matching threshold.
[0043] On the other hand, a voice quality inspection apparatus is provided,
the apparatus
includes quality inspection result obtaining module, the module comprises:
[0044] First obtaining module configured to compare each matching coefficient
of the
keyword type with a preset matching threshold, obtaining quality inspection
result.
[0045] On the other hand, a voice quality inspection apparatus is provided,
the apparatus
includes quality inspection result obtaining module, the module comprises:
4
Date Regue/Date Received 2023-01-17
[0046] First obtaining module configured to set sampling weight for each
keyword type,
obtain sampling coefficient according to the sampling weight and the matching
coefficient of
the keyword type;
[0047] Second obtaining module configured to compare the sampling coefficient
with preset
matching threshold, obtain quality inspection result.
[0048] On the other hand, a computer device, including a memory, a processor
and a
computer program stored in the memory and run on the processor configured to
achieve
following steps when the processor executes the computer program:
[0049] Converting voice data into text data, performing sentence segmentation
processing on
the text data, obtaining text fragments;
[0050] Configuring keyword type according to task type, retrieving keyword
corresponding
to the keyword type, and updating keyword library;
[0051] Selecting keyword text from the keyword library, comparing with the
text fragments,
obtaining matching information of keyword text;
[0052] According to comparison between the matching information of the keyword
text and
the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0053] According to comparison between the matching coefficient of the keyword
type and a
preset matching threshold, obtaining quality inspection result of the task
type.
[0054] In another aspect, a computer readable storage medium stored with a
computer
program configured to achieve following steps when the processor executes the
computer
program:
[0055] Converting voice data into text data, performing sentence segmentation
processing on
the text data, obtaining text fragments;
[0056] Configuring keyword type according to task type, retrieving keyword
corresponding
to the keyword type, and updating keyword library;
[0057] Selecting keyword text from the keyword library, comparing with the
text fragments,
obtaining matching information of keyword text;
[0058] According to comparison between the matching information of the keyword
text and
the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0059] According to comparison between the matching coefficient of the keyword
type and a
Date Regue/Date Received 2023-01-17
preset matching threshold, obtaining quality inspection result of the task
type.
[0060] The above-mentioned voice quality inspection method, apparatus,
computer device
and storage medium, setting keyword type according to task type, selecting
keyword text and
comparing the keyword text with the customer service voice text, obtaining
matching
information and matching coefficient to judge the customer service voice
quality inspection
result, the method avoids the manual quality inspection defects, reduces
enterprise costs,
improves voice quality inspection efficiency, and achieves the systematization
of customer
service performance assessment and customer service satisfaction.
Drawing Description
[0061] Figure 1 is an application process diagram of a voice quality
inspection method in an
embodiment
[0062] Figure 2 is an application environment diagram of a voice quality
inspection method
in an embodiment;
[0063] Figure 3 is a process diagram of a voice quality inspection method in
an embodiment;
[0064] Figure 4 is a process diagram of steps of obtaining text fragments in
an embodiment;
[0065] Figure 5 is a process diagram of steps of updating keyword library in
an embodiment;
[0066] Figure 6 is a process diagram of steps of obtaining matching position
in an
embodiment;
[0067] Figure 7 is a process diagram of steps of obtaining matching
coefficient in an
embodiment;
[0068] Figure 8 is a process diagram of steps of obtaining quality inspection
result in an
embodiment;
[0069] Figure 9 is another process diagram of steps of obtaining quality
inspection result in
an embodiment;
[0070] Figure 10 is a structural diagram of quality inspection result
obtaining module in an
embodiment;
[0071] Figure 11 is a structural diagram of voice quality inspection apparatus
in an
embodiment;
[0072] Figure 12 is an internal structural diagram of a computer device in an
embodiment.
6
Date Regue/Date Received 2023-01-17
Specific embodiment methods
[0073] In order to make clearer application purposes, technical solutions, and
advantages, the
present disclosure is further explained in detail with a particular embodiment
thereof, and with
reference to the drawings. It shall be appreciated that these descriptions are
only intended to be
illustrative, but not to limit the scope of the disclosure thereto.
[0074] The present application provides a voice quality inspection method, the
application
process is as shown in Figure 1. For example, the voice quality inspection
method provided by
this application can be applied to the detection of voice customer service
quality, through
converting voice data 100 into text data 101, passing text data 101 through
matching step 102
to obtain quality inspection result 103, the method can avoid defects of
manual quality
inspection, reduce enterprise costs, improve voice quality inspection
efficiency, and achieve
the systematization of customer service performance assessment and customer
service
satisfaction.
[0075] A voice quality inspection method provided by the application can be
applied in the
application environment as shown in Figure 2. Wherein, terminal 200
communicates with
server 201 through network. For example, the voice quality inspection method
provided by this
application converts voice data into text data and performs sentence
segmentation processing,
configures keyword type according to task type, updates keyword library,
selects keyword text
from keyword library and compares with text fragments, then obtain the
matching information
of keyword text, obtains matching coefficient of the keyword type according to
comparison
between the matching information of the keyword text and the selected keyword
text quantity,
and obtains quality inspection result of the task type according to comparison
between the
matching coefficient of the keyword type and a preset matching threshold.
Wherein, terminal
200 can be but not limited to various personal computer, laptop, smart phone,
tablet computer,
portable wearable device or sub-server, server 201 can be an independent
server or a server
cluster composed of a plurality of servers or cloud computing platform to
achieve.
[0076] In an embodiment, as shown in Figure 3, a voice quality inspection
method is
provided, comprising following steps:
[0077] Si, converting voice data into text data, performing sentence
segmentation processing
7
Date Regue/Date Received 2023-01-17
on the text data, obtaining text fragments;
[0078] S2, configuring keyword type according to task type, retrieving
keyword
corresponding to the keyword type, and updating keyword library;
[0079] S3, selecting keyword text from the keyword library, comparing with the
text
fragments, obtaining matching information of keyword text;
[0080] S4, according to comparison between the matching information of the
keyword text
and the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0081] S5, according to comparison between the matching coefficient of the
keyword type
and a preset matching threshold, obtaining quality inspection result of the
task type.
[0082] Through the above-mentioned steps, the problems of cumbersome operation
steps,
low execution efficiency, poor quality inspection quality, high labor costs
and high subjectively
can be improved when manually sorting out and quality inspecting the voice
communication
records generated between customer service and customer, the method converts
voice data into
text data and compares the text data with the keyword in the keyword library,
sets matching
rule, and obtains quality inspection result, the method can avoid defects of
manual quality
inspection, reduce enterprise costs, improve voice quality inspection
efficiency, and achieve
the systematization of customer service performance assessment and customer
service
satisfaction.
[0083] Because voice data comprises two-way voice record generated by voice
communication between customer and enterprise customer service within a period
of time,
when inspecting voice quality of customer server, it is necessary to
distinguish different voice
content of customer and customer service, there may be noise in voice content,
therefore, it is
impossible to inspect text data that is directly converted into a single text
through voice data,
in step Si, exemplarily, voice data can be converted into text data, then
processing sentence
segmentation on text data, and obtaining text fragments, for example, after
obtaining two-way
voice record generated by voice communication between customers and customer
service
within a period of time, preprocessing the voice data to remove noise and
interference, using
training method such as voice activation detection to separate muted content
and voice content
in the voice data, obtaining time separation tag, recognizing the preprocessed
voice data
through Viterbi algorithm and other voice recognition algorithms to obtain
text data, and using
8
Date Regue/Date Received 2023-01-17
automatic rapid recognition technology and other automatic voice recognition
technologies to
separate roles of text data to obtain customer text data and customer service
text data,
segmenting customer text data and customer service text data according to the
time separation
tag, and adding punctuation marks, segmenting customer text data and customer
service text
data according to color vocabulary of customer text data and customer service
text data, and
adding punctuation marks, querying the word count of text without punctuation
marks in
customer text data and customer service text data, when the text without
punctuation marks
exceeds the preset word count threshold, adding punctuation marks to obtain
text fragments.
[0084] When quality inspecting of voice customer service, in order to select
newer or more
suitable keyword, updating keyword library is required, in step S2,
exemplarily, one or more
task types can be set, then configuring keyword type according to the task
type, retrieving the
meaning of keyword type and common vocabulary corresponding to the application
scenario,
then combining the vocabulary and updating the keyword library, for example,
according to
different task types, such as business consulting, marketing outbound calls,
complaints and
suggestions, etc., configuring different keyword types, retrieving common
vocabularies
corresponding to the meanings of different keyword types, application
scenarios, etc.,
combining the common vocabularies, updating keyword library, adding new
keyword to the
keyword library, and deleting inappropriate, infrequently used or
misclassified keyword in the
keyword library, and updating keyword library.
[0085] In order to inspect whether the voice communication between customer
service and
customer meets the requirements, keyword matching on text data is required, in
step S3,
exemplarily, selecting keyword text from keyword library, comparing with text
fragments and
obtaining matching information of keyword text. For example, according to
different task types
and different keyword types, one or more keyword texts are selected from
keyword library,
performing content matching and traversal to the keyword text from first
clause of text
fragments, if there is a matching between the clause and the keyword text,
recording the serial
number of the clause in text fragments, obtaining the matching position and
recording the
keyword text matching times.
[0086] After obtaining the matching information of the keyword text, counting
the number
of each clause in the text fragments that matches the single keyword text, in
step S4,
9
Date Regue/Date Received 2023-01-17
exemplarily, according to matching information, obtaining matching coefficient
of keyword
text through dividing the maximum number of keyword text with matching
relationship
in each clause by the quantity of selected keyword text.
[0087] After obtaining matching coefficient of keyword text, in order to judge
whether the
voice quality inspection is qualified, in step S5, exemplarily, according to
matching coefficient,
comparing with a preset matching threshold to obtain quality inspection result
of task type, for
example, setting sampling weight for each keyword type according to task type,
the sampling
weight is used to indicate the importance of each keyword type in different
task types, for
example, for the business consulting task type, enterprise can increase the
weight of standard
keyword in the keyword type, for the task type of complaint and suggestion,
enterprise can
increase the weight of forbidden keyword and emotional keyword in the keyword
type, then
comparing with the preset matching threshold according to matching coefficient
and sampling
weight to obtain quality inspection result.
[0088] Before quality inspection of voice communication content of customer
and customer
service, converting voice data into text data, as shown in Figure 4,
converting voice data into
text data, performing sentence segmentation processing on the text data, and
the steps of
obtaining text fragments comprises:
[0089] S11, separating muted content and voice content in voice data,
obtaining time
separation tag;
[0090] S12, segmenting text data according to the time separation tag, adding
punctuation
marks at the end of sentence according to color vocabulary at the end of
sentence;
[0091] S13, querying word count of text without punctuation marks in the text
data, if the text
without punctuation marks exceeds a preset word count threshold, adding
punctuation mark.
[0092] Through the above-mentioned steps, the voice communication content
between
customer service and customer can be divided into muted content and voice
content in a period
of time, simultaneously distinguishing the different voice content of customer
and customer
service, eliminating the noise if existing in the voice content, and adding
punctuation marks to
converted text data to segment sentences to improve the applicability of text
data.
[0093] As shown in Figure 4, in step S11, exemplarily, the method of
separating the muted
Date Regue/Date Received 2023-01-17
content and the voice content in the voice data can use voice activation
detection training
method to detect voice data, when detecting of the mute duration exceeds mute
threshold, such
as 3 seconds, recording the time point, obtaining time separation tag, using
the time separation
tag to separate muted content from voice content, such as 5 seconds, recording
the time point
and obtaining the time separation tag, using time separation tag to separate
muted content from
voice content. Different mute thresholds can be set for different voice
speeds, so that the
separation of muted content and voice content in voice data is more
reasonable, and at the same
time, considering the pause habits of different customers and customer service
during voice
communication, so that the separated voice content is more suitable for later
voice quality
inspection steps.
[0094] As shown in Figure 4, in step S12, exemplarily, when retrieving
vocabulary with
exclamation emotional color at the time separation tag, such as: "ah", "well",
etc., adding an
exclamation mark at the end of sentence, when retrieving vocabulary with
questionable
emotional color at the time separation tag, such as: "what", "why", etc.,
adding a question mark
at the end of sentence, by this way, appropriate punctuation marks can be
added to the text
according to the emotional color of customer or customer service to execute
later voice quality
inspection steps.
[0095] As shown in Figure 4, in step S13, exemplarily, querying the word count
in the text
without punctuation marks in the text data, for each segment in the text data,
if the word count
exceeds the preset word count threshold, such as 20 words, adding a period to
segment the
sentence, otherwise adding a coma to segment the sentence, adding a period to
the last segment
in the text data to segment the sentence, such as 30 words, adding a period to
segment the
sentence, otherwise adding a comma to segment the sentence, and adding a
period to the last
segment in the text data to segment the sentence.
[0096] Before selecting keyword text from keyword library, updating the
keyword library is
required, so that suitable keyword text is selected, as shown in Figure 5, in
some embodiments,
the voice quality inspection method also comprises:
[0097] S21, configuring keyword type according to task type, wherein the
keyword type
includes: standard keyword, forbidden keyword, emotion keyword;
[0098] S22, according to meaning and application scenario of each keyword
type, obtaining
11
Date Recue/Date Received 2023-01-17
corresponding keyword;
[0099] S23, according to the keyword, updating keyword database.
[0100] Through the above-mentioned steps, analyzing different task types,
configuring
different keyword types according to the task type, after further analyzing
the meaning of each
keyword type, combining the keyword type with application scenarios, querying
corresponding
common vocabularies, then forming a keyword library with the common
vocabularies and
updating the keyword library, so as to select appropriate keyword text from
the keyword library
in the future and improve the applicability of the selected keyword text.
[0101] As shown in Figure 5, in step S21, the keyword type configured
according to the task
type includes: standard keyword, forbidden keyword, emotion keyword, for each
keyword type,
one or more single configurations can be set, and one or more keyword texts
can be set in each
single configuration, so as to meet the different considerations and quality
inspection
requirements of enterprises when targeting different task types.
[0102] As shown in Figure 5, in step S22, exemplarily, after configuring each
keyword type,
analyzing the meanings and application scenarios of the keyword types, and
retrieving the
corresponding keywords from the internet. For example, for the task type of
marketing
outbound call, the standard keyword can retrieve the following keywords:
"Hello", "Sir",
"Madam", "Product", "Price", etc., so as to meet the diversity and richness of
keywords
available for different task types.
[0103] As shown in Figure 5, in step S23, exemplarily, after retrieving
keyword and manual
review, performing operations of adding keyword and delete keyword on the
keyword library,
for example, when the keywords of "Product" and "Price" are retrieved, after
manual review
is passed, adding the keywords to the keyword library, querying the keyword
library, if banned
words such as illegal, negative, and negligent are retrieved, after manual
review is passed,
deleting the banned words from keyword library to ensure the integrity of
keyword library, so
as to keep the keyword library in real-time and provide guarantee for later
keyword text
selection.
[0104] After updating keyword library, selecting specific keyword text to form
each single
configuration of each keyword type is required, comparing each single
configuration with text
fragments, as shown in Figure 6, in some embodiments, providing a method of
obtaining
12
Date Regue/Date Received 2023-01-17
matching information of keyword text, comprising:
[0105] S31, according to the keyword library, selecting at least one keyword
text for each
keyword type to form a single configuration of each keyword type;
[0106] S32, performing content matching and traversal from first clause of the
text fragment
with the single configuration, if the first clause matches with keyword text
in the single
configuration, recording serial number of the clause in the text segment, and
obtaining
matching position of the keyword text;
[0107] S33, recording keyword text quantity, wherein the keyword text forms a
matching
relationship between the clause and the single configuration, if the same
keyword text in the
single configuration repeats many times in the same clause, the keyword text
quantity is only
counted once, according to the keyword text quantity, obtaining matching times
of the keyword
text;
[0108] S34, according to the matching position of the keyword text and the
matching times
of the keyword text, obtaining matching inforniation of the keyword text.
[0109] Through the above-mentioned steps, keyword text matching rule in the
voice quality
inspection process can be set in detail and accurately, obtaining and
recording the matching
position and matching times of each single configuration and text fragment,
improving the
convenience of querying keyword matching text information can help analyze the
matching
situation of each keyword text during voice quality inspection process and
improve the
accuracy of voice quality inspection.
[0110] As shown in Figure 6, in step S31, exemplarily, selecting at least one
keyword text
for each keyword type to folin a single configuration of each keyword type,
for example, when
task type is marketing outbound calls, configuring three keyword types:
standard keyword,
forbidden keyword and emotion keyword, for standard keyword, setting three
single
configurations, for forbidden keyword, setting two single configurations, for
emotion keyword,
setting two single configurations, wherein, the keywords selected in single
configuration 1 of
standard keyword are: "Hello", "Sir", "Madam", the keywords selected in single
configure 2
of standard keyword are: "Function", "Price", the selected keyword in single
configuration 3
of standard keyword is: "answer", the selected keyword in single configuration
1 of forbidden
keyword is: "Fraud", the selected keyword in single configuration 2 of
forbidden keyword is:
13
Date Regue/Date Received 2023-01-17
"unclear", the selected keywords in configuration 1 of single emotion keyword
are: "Thanks
you" and "Angry". In this way, to meet the enterprise needs to set different
single
configurations for different keyword types, and the selected keyword texts are
diverse and rich.
[0111] As shown in Figure 6, in step S32, exemplarily, considering the keyword
text in the
single configuration as a whole, starting from the first clause of the text
fragment, performing
content matching and traversal on the single configuration, if the first
clause matches with
keyword text in the single configuration, recording serial number of the
clause in the text
segment, and obtaining matching position of the keyword text, for example,
when the nth
clause in the text fragment is "You are talking nonsense, I don't care what do
you think? It has
nothing to do with me", the selected keyword text of a single configuration of
forbidden
keyword are "whatever", "trash", then considering of a matching relationship
between the nth
clause and the keyword text in a single configuration of the forbidden
keyword, the serial
number of the clause in the text fragment is n, therefore, the matching
position of keyword text
of the single configuration is n, wherein n is a non-negative number.
[0112] As shown in Figure 6, in step S33, exemplarily, recording keyword text
quantity,
wherein the keyword text forms a matching relationship between the clause and
the single
configuration, if the same keyword text in the single configuration repeats
many times in the
same clause, the keyword text quantity is only counted once, for example, a
single
configuration is composed of N keyword texts, and in the N keywords, there are
M keyword
texts having matching relationship with the clause content, and there are Q
keyword texts
appearing K times in the clause content of the M keyword texts, although the P
keyword texts
appears K times in the clause content, the Q keyword texts are only counted
once, considering
that the clause matches the single configuration with only M times, so the
keyword text
matching times of a single configuration is M, wherein, N, M, Q, K are non-
negative integers,
and N>M>Q.
[0113] As shown in Figure 6, in step S34, exemplarily, after obtaining the
matching position
and matching times of a single configuration, combining the matching position
and matching
times, for example, if there is a matching relationship between a clause and a
single
configuration, the serial number of the clause in the text fragment is n, when
a single
configuration is matched with the clause, the matching times is M, then for
the clause, the
14
Date Regue/Date Received 2023-01-17
matching information of the single configuration can be expressed as [n, M],
for another clause,
also there is a matching relationship with single configuration, the serial
number of the clause
in the text fragment is m, when the single configuration matches the clause,
the matching times
is Q, the for the clause, the matching information of the single configuration
can be expressed
as [m, Q], in addition, when there is a matching relationship between a single
configuration
and X clauses, the matching information of the single configuration is
expressed as X*Y, X
represents the quantity of clauses that have matching relationship with a
single configuration,
and Y includes the matching position and matching times obtained when a single
configuration
matches each of X clauses, wherein n, m, M, Q, X and Y are non-negative
integers.
[0114] After obtaining the matching information of single configuration,
because the single
configuration can have a matching relationship with a plurality of clauses,
there are a plurality
of groups of matching positions and matching times, before obtaining the
quality inspection
result, matching information processing is required to obtain more matching
information, as
shown in Figure 7, in some embodiments, an obtaining method is provided:
[0115] S41, according to the matching information of keyword text, matching
times is sorted,
according to the maximum matching times, obtaining the number of maximum
matching times;
[0116] S42, dividing the number of maximum matching times by the selected
keyword text
quantity, obtaining the matching coefficient of a single configuration;
[0117] S43, setting sampling weight for each single configuration, obtaining
sampling
coefficient of single configuration through the sampling weight and the
matching coefficient
of single configuration, the mathematical expression of single configuration
sampling
coefficient sp is:
[0118] sp = w * p
[0119] Wherein, sp is sampling coefficient of single configuration, w is
sampling weight of
single configuration, and p is matching coefficient of single configuration;
[0120] S44, comparing the single configuration sampling coefficient with the
preset
matching threshold of single configuration, obtaining matching coefficient of
keyword type,
the mathematical expression of matching coefficient S of keyword type is:
[0121] S = d(max (0, sp¨t))
d(sp¨t)
Date Regue/Date Received 2023-01-17
[0122] Wherein, S is matching coefficient of keyword type, sp is sampling
coefficient of
single configuration, t is matching threshold of single configuration, max()
means maximum
value, d(-) is differential operator.
[0123] Through the above-mentioned steps, analyzing the specific information
of each
matching position and matching times when each single configuration has a
matching
relationship with different clauses in the text fragment, and selecting an
appropriate number of
matching times for calculation in later step of obtaining the quality
inspection result, so as to
improve accuracy of voice quality inspection result.
[0124] As shown in Figure 7, in step S41, exemplarily, sorting a plurality of
sets of matching
times in the matching information in ascending or descending order, and
obtaining the number
of maximum matching times among the sets, for example, if the expression of a
single
configuration information is { [m, M], [n, N], [q, Q] 1, wherein, M>N>Q,
considering the
number of maximum matching times is M, where n, m, q, N, M, Q are non-negative
integers.
[0125] In step S42, exemplarily, after obtaining the number of maximum
matching times,
dividing the number of maximum matching times by the selected keyword text
quantity,
obtaining the matching coefficient, for example, if the quantity of selected
keyword text by a
single configuration is N, and the number of maximum matching times is M, then
the value of
the matching coefficient p of the single configuration is p = MN, wherein N
and M are non-
negative integers.
[0126] In some embodiments, for step S42, if the keyword type is forbidden
keyword, after
obtaining the number of maximum matching times, subtracting the number of
maximum
matching times from the quantity of keyword texts selected by current single
configuration,
and then dividing the quantity of keyword texts selected by current single
configuration to
obtain the matching coefficient, for example, if the quantity of keyword texts
selected by a
single configuration is N, and the number of maximum matching times is M, then
the matching
coefficient value of single configuration is p = (N - M) IN, wherein N and M
are non-negative
integers.
[0127] In step S43, exemplarily, setting sampling weight for each single
configuration
corresponding to keyword type, the sampling weight is multiplied by the
matching coefficient
of each single configuration and accumulated to obtain the matching
coefficient of the keyword
16
Date Regue/Date Received 2023-01-17
type, for example, the keyword type has i single configurations, the matching
coefficients of i
single configurations are expressed as {pi ,p2 ,...,pi }, and the sampling
weights are set for i
single configurations, expressing as {wl ,w2 then
the calculation formula of the single
configuration sampling coefficient sp is: sp
= =. Pi * Wk
[0128] In step S44, setting single configuration threshold value t, if sp is
greater than t,
considering of a keyword type formed by each single configuration is
qualified, then
calculation foimula of the matching coefficient S of keyword type is: S =
d(max (0, sp¨t))
d(sp¨t)
wherein, k, i are positive integers, max() means maximum value, d() is
differential operator.
[0129] In some other embodiments, for step S43 and step S44, the calculation
method of
matching coefficient of keyword type also comprises:
[0130] For a keyword type, there are i single configurations, the matching
coefficients of the
i single configurations are expressed as {p1 ,p2 and
the sampling weights are set for the
i single configurations expressed as {wl ,w2 ,...,wi}, the sampling
coefficients of the i single
configurations are sp = Ispl ,sp2 ,spil = {pl*wl ,p2*w2
,pi*wi}, setting the threshold
of the i single configurations as t = {t1 ,t2 ,ti},
if the sampling coefficient of the k-th single
configuration is greater than the matching threshold of the k-th single
configuration, the k-th
single configuration is considered to be qualified, when all quality
inspections of i single
configurations are qualitied, considering of the quality inspection of a
keyword type composed
of i single configurations is qualitied, then the calculation formula of
matching coefficient S of
d (max (0, spk¨ tk))
a keyword type is: S = lbc=i , ,
wherein, wherein, k, i are positive integers,
d(SPk¨ tic)
max() means maximum value, d(-) is differential operator.
[0131] After obtaining the matching coefficient of each keyword type,
comparing the
matching coefficient with preset matching threshold of each keyword type is
required, as
shown in Figure 8, in some embodiments, an obtaining method is provided:
[0132] S51, according to comparison between matching coefficient of each
keyword type
and each preset matching threshold, obtaining quality inspection result.
[0133] In step S51, exemplarily, after obtaining matching coefficient of each
keyword type,
the matching coefficient of each keyword type is directly compared with preset
matching
17
Date Regue/Date Received 2023-01-17
threshold of each keyword type, obtaining quality inspection result according
to the comparison
result, for example, there are three keyword types, they are standard keyword,
forbidden
keyword, and emotion keyword, the matching coefficient of the three keyword
types are Si,
S2 and S3, the preset matching thresholds of the three keyword types are Ti,
T2 and T3, when
S1>T1 and S2>T2 and S3>T3, the three keyword types are considered qualified,
and the quality
inspection result is qualified.
[0134] In some embodiments, also providing an obtaining method:
[0135] S52, setting sampling weight to each keyword type, according to the
matching
coefficient of sampling weight and the keyword type, obtaining sampling
coefficient of the
keyword type, the mathematical expression of sampling coefficient SP of the
keyword type is:
[0136] SP = S * W
[0137] Wherein, SP is sampling coefficient of keyword type, S is matching
coefficient of
keyword type, W is sampling weight of keyword type;
[0138] S53, comparing sampling coefficient with a preset matching threshold,
obtaining
quality inspection result.
[0139] In step S52, exemplarily, after obtaining the matching coefficient of
each keyword
type, pre-setting the sampling weight of each keyword type, multiplying the
matching
coefficient of each keyword type with the sampling coefficient of each keyword
type, and then
accumulating them to obtain the sampling coefficients, for example, there are
three keyword
types, they are standard keyword, forbidden keyword, and emotion keyword, the
matching
coefficient of the three keyword types are Si, S2 and S3, the sampling weights
of the three
keyword types are Wl, W2 and W3, the calculation formula of the sampling
coefficient SP is
SP = S1*Wl+52*W2+53*W3.
[0140] Through the above-mentioned steps, the method can meet the requirements
for
comprehensive evaluation and judgment on the quality inspection result of each
task type when
enterprises have different emphasizes on different task types, so as to obtain
voice quality
inspection result with improved accuracy and greater rationality.
[0141] In step S53, exemplarily, after obtaining the sampling coefficient,
comparing the
sampling coefficient with the preset matching threshold to obtain the quality
inspection result,
18
Date Regue/Date Received 2023-01-17
for example, there are three keyword types, they are standard keyword,
forbidden keyword,
and emotion keyword, the matching coefficient of the three keyword types are S
= {S1, S2,
S3}, the sampling weights of the three keyword types are W = {W1, W2, W3}, the
calculation
formula of the sampling coefficient SP is SP = S1*W1+S2*W2+S3*W3, the preset
matching
threshold expression is T, if SP is greater than T, then considering of the
quality inspection
result is qualified.
[0142] In some embodiments, as shown in Figure 9, for step S52, the method of
obtaining
sampling coefficient also comprises:
[0143] After obtaining the matching coefficient of each keyword type, pre-
setting the
sampling weight of each keyword type, multiplying the matching coefficient of
each keyword
type with the sampling coefficient of each keyword type, and then accumulating
them to obtain
the sampling coefficients, keep the vector form of the sampling coefficient of
each keyword
type, for example, there are three keyword types, they are standard keyword,
forbidden
keyword, and emotion keyword, the matching coefficient of the three keyword
types are S =
{S1, S2, S3}, the sampling weights of the three keyword types are W = {W1, W2,
W3}, the
calculation formula of the sampling coefficient SP is SP = {SP', SP2, SP3} =
{Sl*W 1+S2*W2+S3*W3}.
[0144] In some embodiments, for step S53, the steps of obtaining the sampling
coefficient
also comprises:
[0145] After obtaining the sampling coefficient in the vector form, pre-
setting the matching
threshold in the vector form, comparing the sampling coefficient in the vector
form with the
matching threshold in the vector to obtain quality inspection result, for
example, there are three
keyword types, they are standard keyword, forbidden keyword, and emotion
keyword, the
matching coefficient of the three keyword types are S = {S1, S2, S3}, the
sampling weights of
the three keyword types are W = {W1, W2, W3}, the calculation formula of the
sampling
coefficient SP is:
[0146] SP = ISP1, SP2, SP31 = {Sl*Wl+S2*W2+53*W3}, the preset matching
thresholds
19
Date Regue/Date Received 2023-01-17
of the three keyword types are {TI, T2, T3}, if S1>T1 and S2>T2 and S3>T3, the
quality
inspection result is qualified.
[0147] What should be noted is although the steps of the process diagram of
figure are shown
in sequence as indicated by the arrows, these steps are not necessarily
executed in the order
indicated by the arrows. Unless explicitly provided instruction in this
article, there is no strict
order in which these steps can be performed, and they can be performed in any
other orders. In
addition, at least partial steps of figure can include more sub steps or
multiple stages, these sub
steps or stages are not necessarily completed at the same time but can be
executed in different
time, the execution order of these sub steps or stages is also not necessarily
in sequence order
but can be performed alternately with the other steps or sub steps of other
steps or at least one
part of the other stages.
[0148] In an embodiment, as shown in Figure 10, providing a voice quality
inspection
apparatus, the apparatus comprises quality inspection result obtaining module,
the module
comprises:
[0149] First obtaining module configured to set sampling weight for each
keyword type,
obtain sampling coefficient according to the sampling weight and the matching
coefficient of
the keyword type;
[0150] Second obtaining module configured to compare the sampling coefficient
with preset
matching threshold, obtain quality inspection result.
[0151] Through the quality inspection obtaining module, enterprise can analyze
and evaluate
the importance of different keyword types, set sampling weights for each
keyword type
according to different requirements, then pre-setting matching threshold, and
comprehensively
evaluating the matching result of each keyword type.
[0152] In first obtaining module, exemplarily, after obtaining the matching
coefficient of
each keyword type, pre-setting the sampling weight of each keyword type,
multiplying the
matching coefficient of each keyword type with the sampling coefficient of
each keyword type,
and then accumulating them to obtain the sampling coefficient, for example,
there are three
keyword types, they are standard keyword, forbidden keyword, and emotion
keyword, the
matching coefficient of the three keyword types are Si, S2 and S3, the
sampling weights of the
three keyword types are Wl, W2 and W3, the calculation formula of the sampling
coefficient
Date Regue/Date Received 2023-01-17
SP is SP = S1*W1+S2*W2+S3*W3.
[0153] In second obtaining module, exemplarily, after obtaining the sampling
coefficient,
comparing the sampling coefficient with the preset matching threshold to
obtain the quality
inspection result, for example, there are three keyword types, they are
standard keyword,
forbidden keyword, and emotion keyword, the matching coefficient of the three
keyword types
are S {S1, S2, S3}, the sampling weights of the three keyword types are W {W1
,W2 ,W3},
the calculation formula of the sampling coefficient SP is SP =
S1*W1+S2*W2+S3*W3, the
preset matching threshold expression is T, if SP is greater than T, then
considering of the quality
inspection result is qualified.
[0154] In some embodiments, the first obtaining module also comprises: after
obtaining the
matching coefficient of each keyword type, pre-setting the sampling weight of
each keyword
type, multiplying the matching coefficient of each keyword type with the
sampling coefficient
of each keyword type, and then accumulating them to obtain the sampling
coefficient, keep the
vector form of the sampling coefficient of each keyword type, for example,
there are three
keyword types, they are standard keyword, forbidden keyword, and emotion
keyword, the
matching coefficient of the three keyword types are S = {S1, S2, S3}, the
sampling weights of
the three keyword types are W {W1, W2, W3}, the calculation formula of the
sampling
coefficient SP is SP = IS l*Wl+S2*W2+S3*W31.
[0155] In some embodiments, the second obtaining module also comprises: after
obtaining
the sampling coefficient in the vector form, pre-setting the matching
threshold in the vector
form, comparing the sampling coefficient in the vector form with the matching
threshold in the
vector to obtain quality inspection result, for example, there are three
keyword types, they are
standard keyword, forbidden keyword, and emotion keyword, the matching
coefficient of the
three keyword types are S = {S1, S2, S3}, the sampling weights of the three
keyword types are
W = {W1, W2, W3}, the calculation formula of the sampling coefficient SP is:
SP = ISP1, SP2,
SP31 = {S1*W1+S2*W2+53*W3}, the preset matching thresholds of the three
keyword types
are {T1, T2, T3), if S1>T1 and S2>T2 and S3>T3, the quality inspection result
is qualified.
21
Date Regue/Date Received 2023-01-17
[0156] As shown in Figure 11, a voice quality inspection apparatus also
comprises a voice
converting module, a task configuration module, a matching information
obtaining module, a
matching coefficient obtaining module and a quality inspection result
obtaining module,
converting voice data into text data, performing sentence segmentation
processing on the text
data, obtaining text fragments, configuring keyword type according to task
type, retrieving
keyword corresponding to the keyword type, and updating keyword library,
selecting keyword
text from the keyword library, comparing with the text fragments, obtaining
matching
information of keyword text, according to comparison between the matching
information of
the keyword text and the selected keyword text quantity, obtaining matching
coefficient of the
keyword type, according to comparison between the matching coefficient of the
keyword type
and a preset matching threshold, obtaining quality inspection result of the
task type.
[0157] Through the above-mentioned apparatus, the problems of cumbersome
operation
steps, low execution efficiency, poor quality inspection quality, high labor
costs and high
subjectively can be improved when manually sorting out and quality inspecting
the voice
communication records generated between customer service and customer, the
apparatus
converts voice data into text data and compares the text data with the keyword
in the keyword
library, sets matching rule, and obtains quality inspection result, the
apparatus can avoid defects
of manual quality inspection, reduce enterprise costs, improve voice quality
inspection
efficiency, and achieve the systematization of customer service performance
assessment and
customer service satisfaction.
[0158] In some embodiments, the steps of task parameters configuration module
comprises:
[0159] Configuring keyword type according to task type, wherein the keyword
type includes:
standard keyword, forbidden keyword, emotion keyword;
[0160] Retrieving the meaning and application scenario of each keyword type,
obtaining
corresponding keyword;
[0161] According to the keyword, updating keyword database.
[0162] In some embodiments, the steps of quality inspection obtaining module
comprises:
[0163] According to comparison between matching coefficient of each keyword
type and
each preset matching threshold, obtaining quality inspection result.
22
Date Regue/Date Received 2023-01-17
[0164] In some embodiments, the steps of quality inspection obtaining module
comprises:
[0165] Setting sampling weight for each single configuration, obtaining
sampling coefficient
through the sampling weight and the matching coefficient of the keyword type.
[0166] Comparing the sampling coefficient with preset matching threshold,
obtaining quality
inspection result.
[0167] In some embodiments, after the matching coefficient of each single
configuration,
setting sampling weight for each single configuration, setting sampling weight
for each single
configuration corresponding to keyword type, multiplying the sampling weight
by the
matching coefficient of each single configuration and accumulating to obtain
the matching
coefficient of the keyword type, setting the sampling weight of each keyword
type, multiplying
the matching coefficient of each keyword type with the sampling coefficient of
each keyword
type, and accumulating them to obtain the sampling coefficient, then comparing
the obtained
sampling coefficient with preset matching threshold, if the sampling
coefficient is greater than
the matching threshold, considering of the voice quality inspection result is
qualified, otherwise
the result is not qualified.
[0168] In some embodiments, after obtaining the matching coefficient of each
single
configuration of standard keyword type, comparing the matching coefficient of
each single
configuration of standard keyword type with preset threshold value of each
single configuration
of standard keyword type, if the coefficient of each single configuration of
standard keyword
type is greater than the preset threshold of each single configuration of
standard keyword type,
considering of the standard keyword type is qualified, then obtaining the
matching efficient of
each single configuration of forbidden keyword type, comparing the matching
coefficient of
each single configuration of forbidden keyword type with preset threshold
value of each single
configuration of forbidden keyword type, if the coefficient of each single
configuration of
forbidden keyword type is greater than the preset threshold of each single
configuration of
forbidden keyword type, considering of the forbidden keyword type is
qualified, and at last,
obtaining the matching efficient of each single configuration of emotion
keyword type,
comparing the matching coefficient of each single configuration of emotion
keyword type with
preset threshold value of each single configuration of emotion keyword type,
if the coefficient
of each single configuration of emotion keyword type is greater than the
preset threshold of
23
Date Regue/Date Received 2023-01-17
each single configuration of emotion keyword type, considering of the emotion
keyword type
is qualified. If the standard keyword type, forbidden keyword type, and
emotion keyword type
are qualified, considering of the voice quality inspection result is
qualified, if at least one of
the standard keyword type, forbidden keyword type, and emotion keyword type is
unqualified,
the voice quality inspection result is considered unqualified.
[0169] For the specific limitation of a voice quality inspection apparatus can
refer to the
above-mentioned voice quality inspection method, which will not be repeated
here. Each
module of the above voice quality inspection apparatus can be achieved fully
or partly by
software, hardware, and their combinations. The above modules can be embedded
in the
processor or independent of the processor in computer device and can store in
the memory of
computer device in forni of software, so that the processor can call and
execute the operations
corresponding to the above modules.
[0170] In an embodiment, a computer device is provided, the computer device
can be a server
whose internal structure diagram is shown in Figure 12. The computer device
includes a
processor, a memory, a network interface, and a database connected through a
system bus. The
processor of the computer device is configured to provide calculation and
control capabilities.
The memory of the computer device includes non-volatile storage medium and
internal
memory. The memory of non-volatile storage medium has an operation system,
computer
programs and database. The internal memory provides an environment for the
operation system
and computer program running in a non-volatile storage medium. The network
interface of the
computer device is used to communicate with an external terminal through a
network
connection. The computer program is executed by the processor to implement a
voice quality
inspection method.
[0171] The skilled in the art can understand that the structure shown in
Figure 12 is only
partial structural diagram related this application solution and not
constitute limitation to the
computer device applied on the current application solution, the specific
computer device can
include more or less components than what is shown in the figure, or
combinations of some
components or different components to what is shown in the figure.
[0172] In an embodiment, a computer device is provided, including a memory, a
processor
24
Date Regue/Date Received 2023-01-17
and a computer program stored in the memory and ran on the processor
configured to achieve
the following steps when the processor executes the computer program:
[0173] Converting voice data into text data, performing sentence segmentation
processing on
the text data, obtaining text fragments;
[0174] Configuring keyword type according to task type, retrieving keyword
corresponding
to the keyword type, and updating keyword library;
[0175] Selecting keyword text from the keyword library, comparing with the
text fragments,
obtaining matching information of keyword text;
[0176] According to comparison between the matching information of the keyword
text and
the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0177] According to comparison between the matching coefficient of the keyword
type and
a preset matching threshold, obtaining quality inspection result of the task
type.
[0178] In an embodiment, a computer readable storage medium is provided, the
medium
stored with computer program and the processor performs the following steps
when executing
the computer program:
[0179] Converting voice data into text data, performing sentence segmentation
processing on
the text data, obtaining text fragments;
[0180] Configuring keyword type according to task type, retrieving keyword
corresponding
to the keyword type, and updating keyword library;
[0181] Selecting keyword text from the keyword library, comparing with the
text fragments,
obtaining matching information of keyword text;
[0182] According to comparison between the matching information of the keyword
text and
the selected keyword text quantity, obtaining matching coefficient of the
keyword type;
[0183] According to comparison between the matching coefficient of the keyword
type and
a preset matching threshold, obtaining quality inspection result of the task
type.
[0184] The skilled in the art can understand that all or partial of procedures
from the above-
mentioned methods can be performed by computer program instructions through
related
hardware, the mentioned computer program can be stored in a non-volatile
material computer
readable storage medium, this computer can include various embodiment
procedures from the
abovementioned methods when execution. Any reference to the memory, the
storage, the
Date Regue/Date Received 2023-01-17
database, or the other media used in each embodiment provided in current
application can
include non-volatile and/or volatile memory. Non-volatile memory can include
read-only
memory (ROM), programable ROM (PROM), electrically programmable ROM (EPRPMD),
electrically erasable programmable ROM (EEPROM) or flash memory. Volatile
memory can
include random access memory (RAM) or external cache memory. As an instruction
but not
limited to, RAM is available in many forms such as static RAM (SRAM), dynamic
RAM
(DRAMD), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM),
enhanced SRAM (ESDRAM), synchronal link (Synchlink) DRAM (SLDRAM), memory bus
(Rambus), direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and
memory bus dynamic RAM (RDRAM), etc.
[0185] The technical features of the above-mentioned embodiments can be
randomly
combined, for concisely statement, not all possible combinations of technical
feature in the
abovementioned embodiments are described. However, if there are no conflicts
in the
combinations of these technical features, it shall be within the scope of this
description.
[0186] The above-mentioned embodiments are only several embodiments in this
disclosure
and the description is more specific and detailed but cannot be understood as
the limitation of
the scope of the invention patent. Evidently those ordinary skilled in the art
can make various
modifications and variations to the disclosure without departing from the
spirit and scope of
the disclosure. Therefore, the appended claims are intended to be construed as
encompassing
the described embodiment and all the modifications and variations coming into
the scope of
the disclosure.
26
Date Regue/Date Received 2023-01-17