Note: Descriptions are shown in the official language in which they were submitted.
CA 02485547 2004-11-16
64768-378D
SPECIFICATION
DEVICE, METHOD, AND PROGRAM FOR ENCODING l DECODING
OF SPEECH WITH FUNCTION OF ENCODING SILENT PERIOD
This is a divisional of Canadian Patent Appiieatic~n Seriai
Number 2,373,479.
The invention relates to a device for encodingldecoding of digital
information such as a speech signal, in particular, to a technique for
encoding/dec;oding of a voice-less period. ,
Conventionally, some devices are proposed to reduce an
average bit rate of transmission of a speech signal in a voice-less
period (a period with no voice), by encoding a speech signal at lower bi!
rates than that used to encode a speech signal in a period with a voice.
For example, the technique is disclosed in a document 1 (IEEE
Communication Magazine, pages 84 - 73, Sep. 1997).
The conventional encoding device determines whether the input.
signal includes a voice or not, for each frame with a predetermined size,
e.g. 10 milliseconds, and if the signal in the frame includes a voice, the
signal is encoded and decoded in a general speech coding method.
On the other hand, the input signal. includeb no voice, the
conventional coding device discontinuously encodes feature parameters
of the input speech siflnal and transmits the encoded parameters to a
decoding device. Herein, the decoding device smoothes thQ fQaturo
p~ramete~c diccontinuousiy r~ceived, and d~codes a speech signal by
using the smoothed parameters.
CA 02485547 2004-11-16
64768-378D
rA method of determining whether the speech signal is voice-less
or not for each frame, is also disclosed in the document 1. In the
method, a root means square value (hereinafter, referred to as "RMS")
computed from an input speech signal for each frame, an RMS
corresponding to a low frequency region, the number of zero crossing,
and filter coefficients representing spectral envelope characteristics are
used.
The determination is done by comparing these values in each
frame with th~ predetermined thresholds.
A method of encoding a speech signal in a period with voice is,
for exempla, disclosed as CELP method (Code Excited Linear
Prediction Coding method) in a document 2 (ITU-T recommendation
6.729, July. 1995).
The CELP method is disclosed in a document 3 (Code-Excited
Linear Prediction: High GZuality Speech at Very Low Bit Rates (IEEE
Proc. ICASSP-86, pp. 937 - 940, 1985)).
in an encoding process of a conventional coding device, fir9t,
speech signal is inputted frame by frame and is processed with linear
predictive analysis to obtain linear predictive (LP) coefficients
representing spectral envelope characteristics of a speech, and an
excitation signal for driving an LP synth~sis filter corresponding to the
spectral envelope characteristics is derived to be encoded.
l;urther, in an encoding process of the excitation signal, each
frame is divided into subframes and encoding of the excitation signal is
performed for each subframe. Herein, the excitation signal is composed
of a pitch el~ment representing a pitch period of the input signal, a
residual element, and gains of these elements. The pitch element is
denoted as an adaptive codevector which is stored in a codebook,
which is referred to as "adaptive codebook", and includes the past
2
CA 02485547 2004-11-16
64768-378D
excitation signal. The rosidual element is denoted as a multipulse
signal composed of a plurality of pulses.
Also, In a decoding process, tv decode a speech signal, an
excitation signal derived by d~coding the pitch element and the residual
element is fed into a synthesis filter composed of decoded filter
coefficients.
In a method of encoding a speech signal in a voice-less period.
as described in the document 1, first, an RMS and filter coefficients
calculated from the speech are encoded at a coding device. Then, at a
decoding device, a multipuise signal and a random signal are generated
so that a root mean square of a sum of them is equal to the decoded
RMS, and the sum of them is fed tv a synthesis filter composed using
the decoded filter coefficients to decode a speech signal In a voice-less
period.
in a voice-less period, the feature parameters are transmitted
only in frames that characteristics of the signal changes, otherwise
nothing is transmitted. However, information showing whether the
feature parameters is transmitted or not is sent in another way.
When the feature parameters are nvt transmitted, the output
speech signal is decoded by repeatedly using the past transmitted
feature parameters. Smoothed RM5 is used for decoding not to cause
a discontinuity of a waveform of the decoded spEech signal.
Fig. 8 shows a block diagram representing a structure of a
conventional encoding device. Referring to Fig. 8, the encoding device
includes a voice part coding circuit 12, a voice-less part coding circuit
14, a signal determining circuit 16, a switching circuit 18, and a bit
sequence generating circuit 20.
A speech signal is inputted frame by frame, for example, in 10
milliseconds unit by an input terminal ~ 0. The signal determining circuit
3
CA 02485547 2004-11-16
64768-378D
16 determines whether the speech Signal from the input terminal 1 D is a
period with voice or a voice-less period for each frame, and passes the
determining result (VAD determination sign) to the switching circuit 18
and a bit sequence generating circuit ~2o.
The voice part coding circuit 12 encodes the speech signal from
the input terminal 10 for each frame, and passes the encoded signs! to
the switching circuit 18.
The voice-less part coding circuit 14 encodes the speech signal
from the Input tem~lnal 10 for each frame, and passes the encoded
signal to the switching circuit 18. Further, the voice-Iesc part coding
circuit 14 sends determination information (DTX determination sign)
indicating whether the encoded signal is transmitted in the voice-less
period, to the bit sequence generating circuit 20.
The switching circuit 18 operates based on the VAD
determination sign received 'from the signal determining circuit 1fi.
When the circuit 18 receives the sign indicating a voice period, the
encoded signal passed from the voice part coding circuit 12 is sent to
the bit sequence generating circuit 20. On the other hand, when the
circuit 18 receives the sign indicating a voice-less period, the encoded
signal passed from the voice-less part coding circuit 14 is sent to the bit
sequence generating circuit 20.
The bit sequence generating circuit 20 multipl~x~s the VAD
determination sign from the signal determining circuit 16, the DTx
determination sign from the voice-less part coding circuit 10, and
encoded signal from the switching circuit 18, to generate bit sequence
and outputs the bit sequence from an output terminal 22.
Fig. 9 show: a block diagram for explaining a conventional
decoding device.
4
CA 02485547 2004-11-16
64768-378D
Referring to Fig. 9, the decoding device includes a bit sequence
decomposing circuit 28, a switching circuit 28, a voice part decoding
circuit 30, and a voice-less part decoding circuit 34.
The bit sequence decomposing circuit 26 decomposes a bit
sequence inputted from an input terminal 24 into the VAD determination
sign, the DTX determination sign, and the encoded signal. And then.
the circuit 28 sends the VAD determination sign and the encoded signal
to the switching circuit 28, and sends the DTX determination sign to the
voice-less part decoding circuit 34.
The switching circuit 28 operates based on the VAD
determination sign received from the bit sequence decomposing circuit
2B. When the circuit 28 receives the sign indicating a voice period, the
encoded signal passed from the bit sequence decomposing circuit 26 is
sent to the voice part decoding circuit 30. On the other hand, when the
circuit 28 receives the sign indicating voice-less period, the encoded
signal passed from the bit sequence decomposing circuit 26 is sent to
the voice-less part decoding circuit 34.
The voice part decoding circuit 30 decodes the encoded signal
passed from the switching circuit 28 and outputs the decoded signal
from an output terminal 32.
The voice-less part decoding circuit 34 decodes the encoded
signal passed from the switching circuit 28 by using the DTX
determination sign from the bit sequence decomposing circuit 26, and
ou~uts the decoded signal from an output tem~inal 32.
Fig, 10 shows a block die~ram representing a voice-less part
decoding circuit 34 of a conventional decoding device. Referring to Fig.
10, the voice-ie~ part decoding circuit 3~4 includes a parameter
decoding circuit 54, a random circuit 56, a pulse circuit 53, a pitch circuit
58, a mixing circuit B1, a smoothing circuit 66, and a synthesis circuit 88.
CA 02485547 2004-11-16
64768-378D
The parameter decoding circuit 54 decodes filter coefficients and
an RMS from the encoded signal Inputted from an input terminal 52,
and sends the filter coefficients and the RMS to the synthesis circuit 68
and the smoothing circus 66, respectively.
The smoothing oircuit 66 receives th~ RMS from the parameter
decoding circuit 54, and smoothes the RMS. And then the circuit 66
passes the smoothed RMS to the mixing circuit 8i . However, if it is
found that the encoded signal is not transmitted through the DTX
determination sign from an input terminal 50, the circuit 66 calculates
the smoothed RMS by smoothing tho RMS valu~s of the past frames.
Herein, a smoothed RMS P(n) which is used in the n-th frame in
a voice-less period is calculated by using the following equation (1 ) with
the RMS p(n) received in the n-th frame. However, when no encoded
signal is transmitted, the RMS of the previous frame is used in the
equation (~ ) instead of p(n).
P(n) _ (1- a ) ' P(n-1 )+ a ' p(n) ....(1 )
Herein, a is a smoothing factor for determining a degree of
smoothing, in the above-mentioned document 1, a fixed value x.125 is
set. Further, P(-1 ) is equal to zero.
The random circuit 56 generates a random signal and passes
the random signal to the mixing circuit 61. The pulse circuit 53
generates a multipulse signal composing of a plurality of pulses, each of
which has a location and an amplitude determined based on each
random number, and passes the multipulse signal to the mixing circuit
61.
The pitch circuit 58 generates a pitch signal q(i) composed of the
above-mentioned adaptive codevector, and passes it to the mixing
circuit 61. Since a pitch period used to define the adaptive codevector
is not transmitted, a random number is used instead.
6
CA 02485547 2004-11-16
64768-378D
The mixing circuit 61 computes an excitation signal x(i) to be fed
into a synthesis filter by pertorming the linear sum of the random signal
r(i) from the random circuit 56, the muttipulse signal p(i) from the pulse
circuit 53, and the pitch signal q(i) from the pitch circuit 58, and the
result of the eamputetion is sent to the synthesis circuit 68.
A method can be used of computing coupling coefficients of the
linear sum as described in the document 1.
In the method, first, a coupling coefficient of the pitch signal Gq
is selected from a limited range of valu~s according to a random
number.
Next, using the Gq, a coupling coefficient of the multipulse signal
(3p is calculated so that the RMS derived from the linear sum of the
pitch signal and the multipulse signal is squat tv the smoothed RMS.
Usin~ thus calculated Gq and Gp, the linear sum of the pitch
signal and the multipulse signal e(i) is calculated according to thQ
following equation (2).
e(i) s Gq ' q(i) + ~P ' p(i) ....(2)
Furthermore, a coupling coefficient of the linear sum of e(i) and
the random signal r(i). Gr(i) and y , is computed so that the RMS derived
form the linear sum of the e(i) and r(i) is equal to the smoothed RMS.
Herein, as a coupling coefficient of the random signal, a fixed value, ~y
=0.6 is used.
Therefore, the excitation signal to be fed into the synthesis filter,
x(i), is computed according to the following equation (3).
X(I) = Gr ~ IGq ~ q(i) + GP ~ P(I)1 + v ~ r(I) ...(3)
The synthesis circuit 68 decodes the encoded signal by feeding
the excitation signal passed from the mixing circuit 61 to ~ synthesis
filter composed of the filter coefficients passed~from the parameter
decoding circuit 54. Then, the circuit B8 outputs the decoded speech
CA 02485547 2004-11-16
64768-378D
signal from an output terminal 70.
However, the above-mentioned conventional device includes the
following problems.
Th~ first problem is that ther~ may be a case where ftlter
coefficients used to decode a speech signal in a voice-Less period
changes discontinuously at a decoding device, end therefore,
degradafion of a quality of decoded signal occurs.
That reason Is because dlscontfnuously transmitted filter
coefficients are used as they are.
The second problem is that a decoding process in the beginning
period (for example, several hundreds of milliseconds) in a voice-Icss
period may b~ influenced by a voice period right before the voice-less
period, and consequently an amplitude of the decoded signal is
Increased over the actual amplitude or degradation of speech quality of
the decoded signal occurs, for example, due to existence of echoed
sound.
ThAt reason is b~cause a smoothing process of the RMS is
always performed in a voice-less period to prevent decoded
(reproduced) signals in the voice-loss period from being discontinuous.
The third problem is that decoded signal in a voice-less period is
remarkably different from a background noise of input speech signal tn
hosting tha dscodad signal, and as a result, discontinuous auditory
impression is given between tho background noise included in the
voice-less period and a background noise in a voice period.
That reason is because a fixed value is used as a ratio of a pulse
element and a pitch element to a random element, in generating an
excitation signal to be fed into the synthesis filter in a voice-less period.
Therefore, the invention is considering the problems. It is a main
object of the invention to encode a speech signal in a voice-less period
S
CA 02485547 2004-11-16
64768-378D
in a high performance, and to provide a device which realizes a high
coding quality even if an average transmission bit rate is decreased to
encode a speech signal in a voice-less period.
tt is another object of the inv~ntion to provide a decoding device
which can reduce a d~gradation of th~ speech quality due to
discontinuity of the filter coefficients in decoding a speech signal in a
voice-lass period,
DISGLC,~SURE OF THEyj,'~~~,'~~J"'jiON
According to a first aspect of the invention to r~alize the objects,
a speech decoding device is provided, which changes a decoding
operation of a speech signal according to whether the speech signal is
in a voice period or in a voice-less period in each frame, and which
selects feature parameters representing spectral envelope
characteristics of the speech signal to be decoded from the feature
param~tars, smoothes tha select~d featuro parameters In a timo
direction, and decodes the speech signal by using the smoothed feature
parameters.
According to a second aspect of the invention, a speech
decoding device Is provided which changes a decoding operation of a
speech signal according to whether the speech signal is in a voice
period or in a voice-less period in each from~, and which decodes the
speech signal by using a value, which is obtained by smoothing, in a
time direction, at least one of the feature parameters according to an
elapsed time from a time point when a transition occurs from the voice
period to the voice-less period.
According to a third aspect of the invention, a speech decoding
device is provided which changes a decoding operation of a speech
signal according to whether the speech signal is in a voice period or in a
9
CA 02485547 2004-11-16
64768-378D
voice-less period in each frame, and which decodes the voice signal by
using a value, which Is obtained from at least one of the received
feature parameters as it is in a certain time period immediately alter
changing from the voice period to the voice-less period, and obtained by
smoothing et least one of the feature parameters in a time period after
the certain time period.
According to a fourth aspect of the invention, a speech decoding
device is provided which changes a decoding operation of a speech
signal according to wheth~r the spaoch signal is in a voic~ period or in a
voice-less period in each frame, and whioh decodes the speeoh signal
by using a value, which is obtained by smoothing at least one of the
feature parameters according to the feature parameters.
According to a fifth aspect of the invention, a speech decoding
device is provided which changes a decoding operation of a speech
signal according to whether the speech signal is in a voice period or in a
voice-less period in each frame, and which decodes the speech signal
by using a value, which is obtained by smoothing, in a time direction, at
least one of the feature parameters according to at least one of the
feature parameters and an elapse time from when a transition is made
from a voice period to a voice-less period_
According to a fifth aspect of the invention, a speech d~coding
device is provided which changes a decoding operation of a speech
signal according to whether the speech signal is in a voice period or in a
voice-less period in each frame, and which decodes the speech signal
by using a value, which is obtained from at least one of the feature
parameters as it Is when the f~atur~ parameter satisfi~: a
predetermined condition, and obtained by smoothing, in a time direction,
at least one of the feature parameters after the condition is not satisfied.
CA 02485547 2004-11-16
64768-378D
According to a sixth aspect of the invention, a speech decoding
device is provided which changes a decoding operation of a speech
signal according to whether the speech signal is in a voice period or tn a
voice-less period in each frame, and which decodes the speech signal
by using a value which is obtained by smoothing, in a tim~ direction, at
least one of the feature parameters according to an elapsed time from
when a transition is made from a voice period to a voice-less period.
According to a seventh aspect of the Inventlnn, a speech
decoding device is provided which changes a decoding operation of a
speech signal according to whether the speech signal is in a voice
period or in a voice-less period in each frame, end which decodes the
speech signal by using a value, which is obtained from at least one of
the feature parameters as it is when the feature parameter satisfies a
predetermined condition and Immediately after 8 transition is made from
a voice period to a voice-less period, otherwise, obtained by smoothing,
in a time direction, at least one of the feature parameters.
According to an eighth aspect of the invention, a speech
decoding device is provided, which changes a decoding operation of a
speech signal according to whether the speech signal is in a voice
period or In a voice-less period In each frame, and which generates the
a speech signal in a part of a voice-less period by feeding an excitation
signal composed of plural types of signals, and determines coef~oien~
used to perform a sum operation of the plural types of signals oooording
to at least one of the received feature parameters.
According to a ninth aspect of the invention, a speech decoding
device Is provided, which changes a decoding operation of the speech
signal according to whether the speech signal is in a voice period or in a
voice-less period in each frame, and which generates a speech signal in
a voice-less period by feeding an excitation signal composed of plural
11
CA 02485547 2004-11-16
64768-378D
types of signals, and determines, in a part of the period, a
coefficient used to perform a sum operation of the plural
types of signals according to at least one of the feature
parameters smoothed in a time direction.
According to a tenth aspect of the invention, in
the speech decoding device of the above the first aspect to
the ninth aspect, the feature parameter includes at least
one of a quantity representing spectral envelope of the
signal to be decoded and a quantity representing power of
the signals to be decoded.
According to an eleventh aspect of the invention,
a coding device which determines whether the speech signal
is in a voice period or in a voice-less period in each
frame, and encodes a feature parameter of the speech signal
is incorporated with the voice decoding device of the first
aspect to the tenth aspect.
According to a further aspect, there is provided a
speech decoding device which decodes speech signal by using
received feature parameters according to whether the speech
signal is in a voice period or in a voice-less period, the
device comprising: a voice-less part decoding unit which
generates signals in the voice-less period by feeding an
excitation signal composed of plural types of signals to a
synthesis filter in the voice-less period, wherein the
voice-less part decoding unit comprises a weighting
coefficient determining unit which determines a weighting
coefficient used in a weighted sum operation of the plural
types of signals including random signal or multipulse
signal to generate the excitation signal in the voice-less
period according to at least one feature parameter, and the
12
CA 02485547 2004-11-16
64768-378D
excitation signal generated by using the weighting
coefficient is fed to the synthesis filter.
According to another aspect, there is provided a
speech decoding device which decodes speech signal by using
received feature parameters according to whether the speech
signal is in a voice period or in a voice-less period, the
device comprising: a voice-less part decoding unit which
generates signals in the voice-less period by feeding an
excitation signal composed of plural types of signals to a
synthesis filter in the voice-less period, wherein the
voice-less part decoding unit comprises a weighting
coefficient determining unit which determines a weighting
coefficient used in a weighted sum operation of the
plurality of types of signals including random signal or
multipulse signal to generate the excitation signal in the
voice-less period according to at least one smoothed feature
parameter obtained by smoothing a feature parameter selected
from the received feature parameters in a time direction,
and the excitation signal generated by using the weighting
coefficient is fed to the synthesis filter.
According to another aspect, there is provided a
method of decoding speech signal based on received feature
parameters by changing a decoding operation according to
whether the speech signal is in a voice period or in a
voice-lees period, at least a part of decoding process in
the method in the voice-less period comprising the steps of:
determining a weighting coefficient used to generate an
excitation signal of the voice-less period by performing a
weighted sum operation of plural types of signals based on
at least one of the received feature parameters; and
generating the excitation signal based on the weighting
coefficient, and generating speech signal in the voice-less
12a
CA 02485547 2004-11-16
64768-378D
period by feeding the excitation signal to a synthesis
filter.
According to another aspect, there is provided a
method of decoding speech signal based on received feature
parameters by changing a decoding operation according to
whether the speech signal is in a voice period or in a
voice-less period, at least a part of decoding process in
the method in the voice-less period comprising the steps of:
smoothing the received feature parameters and calculating
smoothed parameters; determining a weighting coefficient
used to generate an excitation signal of the voice-less
period by performing a weighted sum operation of plural
types of signals based on at least one of the smoothed
parameters; and generating the excitation signal based on
the weighting coefficient, and generating speech signal in
the voice-less period by feeding the excitation signal to a
synthesis filter.
According to another aspect, there is provided a
recording medium which records a program performing a method
of decoding speech signal based on received feature
parameters by changing a decoding operation according to
whether the speech signal is in a voice period or in a
voice-less period, at least a part of decoding process in
the method in the voice-less period comprising the steps of:
determining a weighting coefficient used to generate an
excitation signal of the voice-less period by performing a
weighted sum operation of plural types of signals based on
at least one of the received feature parameters; and
generating the excitation signal based on the weighting
coefficient, and generating speech signal in the voice-less
period by feeding the excitation signal to a synthesis
filter.
12b
CA 02485547 2004-11-16
64768-378D
According to another aspect, there is provided a
recording medium which records a program performing a method
of decoding speech signal based on received feature
parameters by changing a decoding operation according to
whether the speech signal is in a voice period or in a
voice-less period, at least a part of decoding process in
the method in the voice-less period comprising the steps of:
smoothing the received feature parameters and calculating
smoothed parameters; determining a weighting coefficient
used to generate an excitation signal of the voice-less
period by performing a weighted sum operation of plural
types of signals based on at least one of the smoothed
parameters; and generating the excitation signal based on
the weighting coefficient, and generating speech signal in
the voice-less period by feeding the excitation signal to a
synthesis filter.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1 shows a diagram of a structure of a voice-
less part decoding circuit according to a first embodiment
of the invention.
Fig. 2 shows a diagram of a structure of a
decoding device according to a second embodiment of the
invention.
Fig. 3 shows a diagram of a structure of a voice-
less part decoding circuit according to a second embodiment
of the invention.
Fig. 4 shows a diagram of a structure of a
decoding device according to a third embodiment of the
invention.
12c
CA 02485547 2004-11-16
64768-378D
Fig. 5 shows a diagram of a structure of a voice-
less part decoding circuit according to a third embodiment
of the invention.
Fig. 6 shows a diagram of a structure of a
decoding device according to a fourth embodiment of the
invention.
Fig. 7 shows a diagram of a structure of a voice-
less part decoding circuit according to a fourth embodiment
of the invention.
12d
CA 02485547 2004-11-16
64768-378D
Fig. 8 shows a diagram of a structure of a coding device
according tv a conventional device and the invention.
Fig. 9 shows a diagram of a structure of a conventional decoding
device.
Fig.10 shows a diagram of a structure of a voice-less part
decoding circuit of a conventional decoding device.
BEST MODE FOR F_Mi~ODYING THE INVFNT'ION
D~scription is made about embodiments of the invention. A
speeoh deoodlng device acoording to a first embodiment of the
invention includes a switching device (shown in Fig. 9 (28)), a
smoothing device (shown in Fig. 1 (84)), and a group of decoding
devices (shown in Fig. 1 (56, 53, 58, 61, and 68)).
The switching device switches the method of decoding the signal
by using the feature parameters of the Qncoded signal to be dQCOdQd,
according to determination information representing whether the
encoded signal is in a voice period or in a voice-less period for each
frame. The smoothing device smoothes the feature parameters
representing spectral envelope characteristics of the encoded signal.
The group of decoding devices decodes the encoded stgnat by using
the smoothed feature parameters.
A speech decoding device according to a second embodiment of
the invention includes a switching device (shown in Fig. 2 (28)), a group
of smoothing devices (shown in Fig. 2 (36) and Fig. 3 (49 and 51 )), and
a group of decoding devices (shown in Fig. 3 (56, 53, 58, 61, and 68)).
The switching device switches the method of decoding the signal
by using th~ feature parameters of encoded signal to be d~coded,
according to determination information representing whether the
encoded signal is in a voice period or in a voice-less period for each
13
CA 02485547 2004-11-16
64768-378D
frame. The group of smoothing devices smoothes at least one
parameter in the feature parameters, based on the parameters and en
elapsed time from a time point when a voice period is changed to a
voice-less period. The group of decoding devices decodes the encoded
signals by using th~ smoothed feature parameters.
A speech decoding device according to a third embodiment of
the invention includes a switching device (shown in Fig. 2 (28)), a group
of smoothed value generating devices (shown in Fig. 2 (3f ) and Flg. 3
(49 and 51 )), and a group of decoding devices (shown in Fig. 3 (56, 53,
58, 61, and 68)).
The switching device switches methods of decoding the signal
by using feature parameters of encoded signals to be decoded,
according to determination information representing whether the
encoded signal Is In a voice period or In a voice-less period for each
frame. The group of smoothed value generating devices set the original
value of at least one of transmitted f~ature parameters as a smoothed
value immediately after transition from a voice period to a voice-less
period and when a feature parameter satisfies predetermined conditions,
and thereafter, generate a smoothed value by smoothing at least one of
the feature parameters. The group of decoding devices decodes the
encoded signals by using the smoothed parameters.
A speech decoding devise aoeordlng to a fourth embodiment of
the invention includes a switching device (shown in Fig.. 4 (28)), a group
of signal generating devices (shown in Fie. 5 (56. 53, 58. 60. and 68)),
and a coefficient determining device (shown in Pig. 5 (38)).
The switching device switches the method of decoding the signal
by using the feature parameters of encoded signals to be decoded,
according to determination information representing whether the
encoded signal is in a voice period or in a voice-less period for each
14
CA 02485547 2004-11-16
64768-378D
frame. The group of signal generating devices generates a decoded
signal of a voice-less period by feeding an excitation signal composed
of plural types of signals into a synthesis filter. The coefficient
determining device determines coefficients used to mix plural types of
signals in the voice-less period according to at least one of the received
feature parameters.
A speech decoding device according to a fifth embodiment of the
Invention Includes a switching device (shown In I=Ig. 6 (28)), a group of
signs) generating devices (shown in Fig. 7 (56, 53, 58, 62, and 68)), a
group of parameter oaloul~ting devioes (shown in Fig. 7 (49 and 51 ),
end a coefficient determining devioe (shown in Fig. 6 (38)).
The switching device switches methods of decoding signals by
using feature parameters of encoded signals to be decoded, according
to determination Information representing whether the encoded signal Is
in a voice period or in a voice-less period for each frame. The group of
signal generating devices generates a signal of a voice-less period by
feeding an excitation signal composed of plural types of signals into a
synthesis filter. The group of parameter calculating devices calculates a
smoothed parameter by smoothing the received feature parameters.
The coefficient determining device determines coefficients used to mix
plural types of signals in the voice-less period according to at least one
of the calculated feature parameters.
In a speech decoding device according to a sixth embodiment of
the invention, the feature parameters include at least one of a value
representing the spectral envelope of the signals to be decoded and a
value representing a power of the signals.
A preferred embodiment of a encodingldecoding device
according to the invention includes a encoding device (shown in Fig. 8)
which determines whether the input signal is in a voice period or in a
CA 02485547 2004-11-16
64768-378D
voice-Icss period for each frame and encodes feature parameters of the
input signal, and a speech decoding device according to one of the
devices shown in the first embodiment to the sixth embodiment.
Description is made about an operation and a principle of an
embodiment of tho invention.
According to the invention, the speech decoding device
smoothes a discontinuously transmitted filter coefficients with the RMS,
end uses the coefficients about a synthesis filter, in decoding a speech
signal in a voice-less period. Thereby, a discontinuous change of the
falter coefficients can be prevented which is caused due to the
discontinuous transmission of the filter coefficients, and as a result, a
voice quality of the decoded signal can be improved.
In the speech decoding device, when the filter coefficients and
the RMS which are smoothed In a voice-less period are currently used,
the filter coefficients and the RMSs of the past frames influence the
currently used filter coefficients and the RMS because of the smoothing
process,
Since the signal in the beginning of the voice-less period
includes characteristics of a voice period immediately before the voice-
less period, the signal in the voice-less period is decoded by using the
feature parameters Including the characterlstlcs of the voice period.
Consequently, an amplitude of a waveform of the decoded signal
become larger than an actual amplitude of the input speech signal, or
degradation of the decoded speech signal, such as an existence of
echo in the decoded signal, may occur.
To prevent them, when a predetermined time elapses or a
certain number of frames are received from a time point of the transition
from a voice period to a voice-less period, for example, a smoothing
factor is set not to perform smoothing process when a value of the RMS
16
CA 02485547 2004-11-16
64768-378D
representing an amplitude of the decoded speech is still larger than a
predetermined value. Thereby, In the beginning of the voice-less period,
an effect from the voice period immediately before the voice-less period,
due to smoothing of the feature parameter can be reduced.
There may be the auditory difference between a background
noise included in the signal decoded in a voice part decoding circuit and
the signal decoded in a voice-less part decoding circuit, in a case where
background noises are included in the input signal. This reason is that
the voice-less part decoding circuit computes an excitation signal to b~
fed into a synthesis filter, on only condition that the RMS of the signal
becomes equal to a smoothed value of the transmitted RMS.
In the invention, it is capable of reducing degradation of the
decoded speech quality due to the auditory difr~erence, by determining
how to compute the excitation signal considering characteristics of the
input signal. To consider the characteristics, for example, a random
noise signal is mainly used when the smoothed RMS is small, on the
other hand, a pulse signal or a pitch signal is mainly used when the
smoothed RMS is large or when the spectrum computed from the filter
coefficients are not flat.
Description is made in more detail about embodiments of the
invention with reference to the drawings. A basic structure of an
eneodlng device used in the embodiments is similar to the structure of
the coding device shown in Fig. 8. Also, a basic structure of the
de~ding~ device is similar to the structure of the decoding device shown
In Flg. 9.
Fig. 1 shows a block diagram of a structure of a voice-less part
decoding eirouit in a decoding device according to the first embodiment
of the invention. Referring to Fig. 1, the voice-less part decoding circuit
of the first embodiment is different from the voice-less part decoding
17
CA 02485547 2004-11-16
64768-378D
oirouit 34 shown in Fig. 10 in that the former voles-less pert decoding
circuit further includes a smoothing circuit 64. In the following
description, it is mainly explained about the difference between the
device according to the invention and the conventional devic~, therefore,
explanation about common parts will be omitted.
A parameter decoding circuit 54 determines the filter coefficients
and the RMS by using a sequence of signals revived from an input
terminal 52, and passes the determined I<Iter coetflclent and the
determined RMS to the smoothing circuit 64 and the other smoothing
circuit 66, respectively.
The smoothing alrcuit 64 smoothes the filter coeffiolents received
from the parameter decoding circuit 54 and passes the smoothed filter.
coefficients to the synthesis circuit 88. However, the smoothing circuit
64 performs smoothing process by using the filter coeiflclents of the
past frames when the DTX determination sign received Pram an input
terminal 50 indicates that the featur~ param~tors are rec~ivod.
Smoothed filter coefficients F(n, i), (i ~ 1,..., M)~used for the n-th
frame from the beginning of each voice-less period, is calculated by
using an equation (4) with the filter coefficients f(n, i) (i = 1. .... M)
entered in the n-th frame. Also, In a frame where nothing is transmitted,
the filter coefficients sent immediately before the frame arse used to
calculate instead of f (n, i).
F (n, i) _ (1 - ~ )F(n-~. i) + ~B f(n, i) ...(4)
Herein, ~i is a smoothing factor to determine a degree of
smoothing, Also, F (-1, i). (i = 1. ..., M) is equal tv 0.
M is an order of the synthesis (liter. The synthesis circuit 88
decodes the signal by feeding an excitation signal receiv~d from the
mixing circuit 61 into the synthesis fitter composed of the filter
coefficients received from the smoothing circuit 64, and outputs the
18
CA 02485547 2004-11-16
64768-378D
decoded signal to an output tarmlnal 70.
Fig. 2 shows s diagram representing a structure of the decoding
devic~ according to thQ second embodiment of the invention. The
embodiment differs from the conventional deeodlng device shown in Fig.
9 in that a structure of a voice-less part decoding circuit 35 of the
embodiment is different from that of the conventional decoding device,
and the embodiment includes a smoothing control circuit 36.
Hereinafter, descrip~on is mainly made about the difference between
the decoding dev)ce according to the second embodiment and the
conventional decoding device, and explanation about parts each of
which is the same as the corresponding pArt of the conventional
decoding device may be omitted for the sake of convenience.
A bit s~quenc~ decomposing circuit 26 decompos~s a bit
sequence supplied from an Input terminal 24 into a VAD determination
sign, a DTX determination sign, and a sequenve of the encoded signal,
and passos tho VAD dotermlnatlon sign to a smoothing control circuit
36 and a switching circuit 28, passes the sequence of the signal to the
switching circuit 28, and passes the DTX determination sign to a voice-
less part decoding circuit 35.
The switching circuit 28 passes the sequence of the signal
passed from tho bit sequence d~composlng circuit 26 to a volco part
decoding circuit 30 when the VAt7 determination sign from the b~
sequence decomposing circuit 26 indicates that the input signal is in a
voice period, or passes the sequence of the signal to a voice-less part
decoding circuit 35 when it indicates that input signal is in a voice-less
period.
The smoothing control circuit 3G passes smoothing factors ~ (n)
and /3 (n) dQtermined based on a changQ of th~ VAD determination sign
from the bit sequence decomposing circuit 26. to the voice-less part
19
CA 02485547 2004-11-16
64768-378D
d~coding circuit 35. Heroin, n rsprosants a framo numbor, countQal
from the beginning, of frames in each voice-less period.
For example, when the VAD detem~ination sign indicates that the
input signal is in a voice-less period, an effect of a part in a voice period
immediately before the voice-less period on the beginning part in the
voice-less period can be r~duced by setting each of values of the
smoothing factors c~ (n) and S (n) to 1 In the first specified frames or for
a specified period in the voice-less period. Further, by setting each of
values of the smoothing factors cx (n) and s (n) tv 1 while a similarly
transmitted parameter such as the filter coefficients or the RMS satisfies
a spQCffiod condition, an affect of a part In a volcQ period immediately
before the voice-less pe~od on the beginning part in the voice-less
period oen be reduoed.
For example, the specified condition is that the RMS is more
than a threshold value or that both the RMS and the RMS of the first
Subirame in the voice-fens period are less than a threshold value, for
detecting that the RMS is under the influence of the part, in a voice
period, immediately b~for~ the voice-less period. Also, the specked
condition may be that a distant (for ex~mple, square distanco)
between the filter coefficients and a predetem~ined filter coefficients is
less than a predetermined threshold value for detecting that the fitter
coefficients are similar to a smoothed spectrum in a voice period.
Further, when a voice period immediately befor~e~ a first voicQ-
less period does not include a certain number of frames or is shorter
than a certain length of period, a smoothed value in the last frame of a
second voice-less period immediately before the voice period can be
used as an Initial value P(-1 ), F(-1, i), (i = 1, ..., M) for calculating
smoothed values of thQ flltQr coefficients and the RMS, since it is
considered that the characteristics of the input signal in the seoond
ao
CA 02485547 2004-11-16
64768-378D
voice-less period is similar to the characteristics of the input signal fn
the first void-less period.
The voice-less part decoding circuit 35 decodes the signal tn a
voice-less period by using the smoothing factors a (n) and (3 (n), the
DTX determination sign received from the bit sequence decomposing
circuit 26, and the sequence of the signal received from the switching
circuit 28, and outputs the decoded signal to an output terminal 32.
Flg. 3 shows a diagram representing a Structure of the voice-less
part decoding circuit 35 according to the second embodiment of the
invention. The voice-less part decoding circuit 36 is different from the
voice-part decoding circuit of the first embodiment of the invention in s
structure of a smoothing circuit 49 and a smoothing circuit 51.
A parameter decoding circuit 54 determines the flter coefficients
and the RMS based on a sequence of.the encoded signal entered from
an input terminal 52, and passes the filter coefficients to tho smoothing
circuit 49 and passes the RMS to the smoothing circuit 51.
The smoothing circuit 49 smoothes the filter coefficients supplied
from the parameter decoding circuit 54 by using a smoothing factor ~
(n) entered from an input terminal 85, and passes the smoothed filter
coefficients to a syntheses circuit 68. However, when the DTX
determination sign received from an input terminal 50 indicates that the
encoded signal is not transmitted the filter coefficients of the previous
frem~ is repeatedly usod.
The smoothed filter coefficients used in the n-th frame from the
beginning of each voice-less period, F (n. i), (i = 1, ..., M) can be
calculated by using the following equation (5) which is similar to the
above equation (4), with the filtQr coafficionts onterod in tha n-th frame
f(n, i).
F(n, i) ~ ('t-~ (n)) ~ F(n-1, i) + ~ (n) - f(n, i) ...(5)
21
CA 02485547 2004-11-16
64768-378D
Heroin, a value of p (n) is changed according to the number of
frames which have already received in each voice-less period, and
takes about 1 when a few frames are received, so as to remove an
effect from the past frames. For example, it can be set as follows.
~ (1 )_ ~3 (2)=1.0, ~ (3)= a (4)=..,_ ~ (L)=0.7. Herein, L is the
number of frames in each voice-less period.
The smoothing circuit 51 smoothes the RMS sent from the
parameter decoding circuit 54 and passes the smoothed RMS to a
mixing circuit 61. However, when the DTX determination sign sent from
an input terminal 60 indicates that the encoded signal ie not transmitted,
a smoothing process is performed by using the RMS recently received.
The smoothed RMS P(n), which is used in the n th frame from the
beginning of each voice-less period, is calculated by using the following
equation (6) which is similar to the equation (1 ), with the RMS p(n)
entered In the n-th frame.
P(n) _ (1- a (n)) ' P(n-1 ) ~ r~ (n) ' p(n) ...(6)
Herein, similarly to j3 (n), c~ (n) is ch~nged ~ccording to the
number of frames which have already received in each voice-less
period, and takes about 1 when a few frames are roaeived, so as to
remove an effect from the past frames. For example, it can be set as
follows.
oc (1 )= cu (2)=1.0, c~ (3)= a (4)=...= a (L)=0.7. Herein, L is the
number of frames in eeoh voice-less period.
Also, one of the processes of the smoothing circuits 49 and 51
can be pertormed. In this case, the filter coefficients or the RMS sent
from the parameter decoding circuit 54 are or is directly sent to the
synthesis circuit 68 or a mixing circuit 61.
In the mixing circuit 61, calculates an excitation signal x(i) to be
fed Into a synthesis filter by performing the linear sum about a random
22
CA 02485547 2004-11-16
64768-378D
signal r(I) sent from a random circuit 66, a pulse signal p(i) sent from a
pulse circuit 53, and a pitch signal 4(i) sent from a pitch circuit ~8 with a
smoothed RMS sent from the smoothing circuit S9 , and passes the
calculated signal to the synthesis circuit 68.
The synthesis circuit 68 decodes the speech signal by feeding
the excitation signal sent from the mixing circuit 61 into the synthesis
filter composed of the filter coefficients sent from the smoothing circuit
49, and outputs the decoded speech signal from an output terminal 70.
Fig. 4 shows a diagram representing a structure of a decoding
device according to the third embodiment of the invention. The
embodiment differs from the conventional decoding device in a voice-
less part examining circuit 38 and a voice-less part decoding circuit 37.
A bit sequence decomposing circuit 26 decomposes a bit
sequence supplied from an input terminal 2d into a VAD determination
sign, a DTX determination sign, and a sequence of signals, and passes
the VAD determination sign and the sequ~nce of signals to a switching
oirouit 28, and passes the DTX determination sign to a voioe-less part
decoding circuit 37.
The switching circuit 28 passes the signal passed from the bit
sequence decomposing circuit 26 to a voice part decoding circuit 30
when the VAD determination sign from the bit sequence decomposing
circuit 26 indicates that the input signal is in a voice period, or passes
the sequence of signals to a voice-less part deoading oirauit 37 when it
indicates that the input signal is in a voice-less period.
The voice-less part examining circuit 38 determines a set up
parameter to adJust coupling coefficients of the linear sum used at the
mixing circuit 62 shown in F ig. 5 by using the filtor coetficiQnts and the
RMS sent from the voice-less part decoding circuit 37, and passes the
parameters to the voice-less part decoding circuit 3?. Description will
23
CA 02485547 2004-11-16
64768-378D
bo made later with a process in the mixing circuit G2 about oalaulation of
the set up parameters.
>=tg. 5 shows a diagram representing a structure of the voice-less
part decoding circuit 37 according to the third embodiment of the
invention. The voice-less part decoding circuit 37 is different from the
voice-part decoding circuit 35 of the first embodiment of the invention in
a mixing circuit 62 and an output destination of a parameter decoding
cln:uit 54. Hereinafter, description is made mainly about the difference,
and description about the common part is omitted.
A parameter decoding circuit 54 determin~s the filt~r eoeffielents
and the RMS based on a sequence of signals entered from an input
terminal 52, and passes the filter coefficients to the smoothing cirwit 64
and an output terminal 23, and passes the RMS to the smoothing circuit
66 and an output terminal 25.
The smoothing circuit 66 smoothes the RMS passed from the
parameter decoding circuit 54 and passes the smoothed RMS to a
mixing circuit 62. When the DTX determination sign sent from an input
terminal 50 indicates that the encoded signal is not transmitted, the
RMS, which is transmitted immediately before the current frame, is used
to smooth. Further, It can be controlled not to update the smoothed
RMS by setting smoothing factors a (n) and ~3 (n) to zero.
A random circuit 6fi generates a random number and passes the
random number to the mixing oircult 62,
A guise circuit 53 generates a pulse signal composed of a pulse
having a location and an amplitude generated base on the random
number, and passes the pulse signal to the mixing circuit 62.
Th~ mixing circuit 62 calculates coupling coefficients of the
above-mentioned linear sum by using the set up parameter received
from an input terminal GO and the smoothed RMS received from the
24
CA 02485547 2004-11-16
64768-378D
smoothing circuit 66.
Also, the circuit 82 calculates a linear sum signal of the random
signal from the random circuit 56, the pulse stgnat from th~ pulse clrcutt
53, and the pitch signal from the pitch circuit 58 by using the coupling
coefficients, and passes the linear sum signal to the synthesis eirouit 68,
The synthesis circuit 68 decodes input signal by feeding an
excitation signal sent frvrn the mixing circuit 82 into a fitter composed of
the filter coefficients sent from the smoothing circuit 64, and outputs the
decoded signal from an output tQrminal 70.
Next, description is mado about the voice-less pert examining
circuit 38 end the mixing circuit 62.
The voice-less part examining circuit 38 determines the
characteristics of a background noise in a voice-less part, and changes
a calculation method of the coupling coefficients of the pitch signal, the
pulse signal, and the random signal in the mixing circuit, according to
the det~rmined characteristics. As set up parameters to be changed,
there arc an order to decide the coupling coefficients or a coupling
coefficient y .
The voice-less part examining circuit 38 uses information, for
example, the RMS and the filter coefficients to determine the
characteristics of the background in the voice-less part.
According to a method of controlling the set up parameters
based on the above the illustrated information, when the RMS is less
than a predetermined threshold value and thereby it is presumed that
there Is no background noise, or when it is presumed that the input
signal is a white noise since an inclination of spectrum of the input
signal calculated from the filter coeffcients is flat, a contribution rate of
the random signal is expanded, It means that a value of y is reduced
with keeping the order of calculation of the coupling coefficients.
CA 02485547 2004-11-16
64768-378D
Also, the set up parameters of the voice-less period can be
included in a sequence of signals and transmitted with the signals.
Ffg. 6 shows a diagram representing a structure of a decoding
device according to the fourth embodiment of the invention. The
embodiment differs from the second embodiment of the invention in a
voice-less part examining circuit 38 and a voice-less part decoding
circuit 39.
A bit sequence decomposing circuit 26 decomposes a bit
sequence supplied from an input terminal 24 into a VAD determination
sign, a DTX determination sign, and a sequence of signals, and passes
the VAD determination sign to a smoothing oontrol airouit 38 and a
switching circuit 28, passes the sequence of signals to the switching
circuit 28, and passes the DTX determination sign to a voice-less part
decoding circuit 39.
The switching circuit 28 passes the sequence of signals passed
from the bit sequence decomposing circuit 26 to a voice part decoding
circuit 30 when th~ VAD determination sign from the bit sequence
decomposing circuit 2G indicates that the encoded signal is in a voice
period, or passes the sequence of signals to a voice-less part decoding
circuit 39 when It indicates that Input signal is in a voice-less period.
The smoothing control circuit 36 passes the smoothing factors a
(n) and fi (n) which are determined according to a change of the VAD
determination sign sent from the bit sequence decomposing circuit 26 to
the voice-less part decoding circuit 39.
The voice-less part examining circuit 38 determines a set up
parameter to adjust coupling coefflctents of the linear sum used at the
mixing circuit 62 shown in Fig. 7 by using a smoothed RM5 sent from
the voice-less pert decoding circuit 39, and peseta the parameters to
the voice-less part decoding circuit 39.
26
CA 02485547 2004-11-16
64768-378D
The voice-less part detecting circuit 39 can pertorm a set up
parameter determining process by replacing RMS with smoothed RMS
in above-mentionQd process of the voice-less part examining circuit 38.
The voice-less part detecting circuit 38 decades an input signal
in a voice-less period, by using the DTX determination sign from the bit
sequence decomposing circuit 28. the encoded signal from the
switching circuit 28, the smoothing factors ~ (n) and ~i (n) from the
smoothing control circuit 36, and the sQt up parameters from thQ voice-
less part examining circuit 38, and outputs the decoded signal from an
output terminal 32.
Also, smoothed RMS calculated by a smoothing circuit 51 shown
in Fig. 7 and smoothed filter coefficients calculated by a smoothing
circuit 49 ate passed to the voice-less part examining circuit 38.
Fig. 7 shows a diagram representing a struckure of the voice-less
part decoding circuit 39 according to the fourth embodiment of the
invention. The voice-Icss part decoding circuit 39 is different from the
voice-part decoding circuit of the second embodiment of the invention in
that in the fourth embodiment, an output from a smoothing circuit 51 is
supplied to an output terminal 69 and a smoothing circuit 49 is supplied
to an output terminal 63.
In each of the above described embodiments of the invention, a
pitch signal, a pulse signal, end a random signal is used to compute en
excitation signal of a synthesis filter, but any of them can be omitted.
A decoding device according to the invention and a coding
device described in a background section of the specification can be
applied to a radio terminal or a radio base station thereby, a radio voice
communication system using o speech signol compressing technique
can be easily established. Further, a voice terminal can be easily
constructed by storing a program tv perform the above described
27
CA 02485547 2004-11-16
647 68-378D
decoding method of the invention into a storage medium such as a
floppy disk and by loading the program into a personal computer to
which a loudspeaker is connected.
As described above, according to the Invention, the foilowing
effects are obtained.
A first effect of the invention is that speech quality degradation
due to discontinuous change of the fitter coefficients used in decoding
the signal In a voice-less period can be prevented In the decoding
dQVICQ of the invention.
This reason is that the discontinuously transmitted filter
coefficient is smoothed end used in the invention.
A socond effect of the invention is that a speech quality
degradation due tv influence of a voice period immediately before a
voice-less period on the beginning of the voice-less period can be
reduced in the decoding device of the Invention.
This reason is that a smoothing factor is adjusted not to smooth
the feature parameters in the beginning of a voice-less period.
A third effect of the invontion is that auditory discontinuity ceuxd
by a transition between a voice period and a voice-less period can be
reduced In the decodin~ device of the Invention.
This reason Is that when an excitation signal of a reproduction
filter is generated in a voice-less period, ratio of a random element to a
pulse element and a pitch element is oh~nged according to a nature of
input signals.
28
