DESCRIPTION
PITCH CYCLE SEARCH RANGE SETT ING APPARATUS AND PITCH CYCLE
SEARCH APPARATUS
Technical Field
The present invention relates to a pitch cycle search
range setting apparatus and pitch cycle search apparatus,
and more particularly to a pitch cycle search range setting
apparatus and pitch cycle search apparatus used in a CELP
(Code Excited Linear Prediction) type speech encoding
apparatus.
Background Art
In such fields as packet communication typified by
digital communication and Internet communication, or
speech storage, speech signal encoding/decoding
technology is essential for making efficient use of radio
wave transmission path capacity and storage media, and
many speech encoding/decoding methods have been developed
to date.
Among these, a CELP ( Code Excited Linear Prediction)
type speech encoding/decoding method is widely used as
a mainstream method when encoding/decoding speech signals
at a medium or low bit rate. A CELP type speech
encoding/decoding method is disclosed in Document 1 (Proc .
ICASSP '85, pp.937-pp.940, 1985).
In a CELP type speech encoding/decoding method, a
digitized speech signal is divided into frames of
approximately 20 ms, linear predictive analysis of the
speech signal is performed every frame and the linear
predictive count and linear predictive residual vector
are found, and this linear predictive count and linear
predictive residual vector are encoded/decoded
individually. This linear predictive residual vector is
also called an excitation signal vector.
A linear predictive residual vector is
encoded/decoded using an adaptive code book that holds
drive sound source signals generated in the past and a
fixed code book that stores a specific number of fixed- form
vectors (fixed code vectors).
This adaptive code book is used to represent a cyclic
component possessed by a linear predictive residual
vector. On the other hand, the fixed code book is used
to represent a non-cyclic component in a linear predictive
residual vector that cannot be represented with the
adaptive code book. In general, linear predictive
residual vector encoding/decoding processing is
performed in subframe units resulting from dividing
frames into shorter time units (of approximately 5 ms
to 10 ms).
With CELP, the pitch cycle is sought from a linear
predictive residual vector, and coding is performed. A
conventional linear predictive residual pitch cycle
search apparatus is described below. FIG.l is a block
diagram showing the configuration of a conventional pitch
cycle search apparatus.
The pitch cycle search apparatus 10(in FIG.10) is
mainly composed of a Pitch Cycle Indicator (PCI) 11,
Adaptive Code Book 12 (ACB) , Adaptive Sound Source Vector
Generator (ASSVG) 13, Integral Pitch Cycle Searcher
(IPCS) 14, Fractional Pitch Cycle Adaptive Sound Source
Vector Generator (FPCASSVG) 15, Fractional Pitch Cycle
Searcher (FPCS) 16, and Distortion Comparator (DC) 17.
The Pitch Cycle Indicator (PCI) 11 sequentially
indicates to the Adaptive Sound Source Vector Generator
(ASSVG) 13 desired pitch cycles T-int within a preset
pitch cycle search range. For example, when the CELP
speech encoding/decoding apparatus performs encoding and
decoding of a 16 kHz speech signal, and the target vector
pitch cycle search range is preset from 32 to 267 at
integral accuracy, and from 3 2 + 1/2, 33+1/2, ..., to 51 + 1/2
at 1/2 fractional accuracy, the Pitch Cycle Indicator
(PCI) 11 outputs 236 kinds of pitch cycle T-int (T-int
= 32, 33, ..., 267) to the Adaptive Sound Source Vector
Generator (ASSVG) 13. The Adaptive Code Book 12 (ACB)
stores drive sound source signals generated in the past.
Next, the Adaptive Sound Source Vector Generator
(ASSVG) 13 extracts from the Adaptive Code Book 12 (ACB)
the adaptive sound source vector p(t-int) that has
integral-accuracy pitch cycle T-int received from the
Pitch Cycle Indicator (PCI) 11, and outputs it to the
Integral Pitch Cycle Searcher (IPCS) 14.
The processing for extracting adaptive sound source
vector p(t-int) that has integral-accuracy pitch cycle
T-int from the Adaptive Code Book 12 (ACB) is described
below. FIG.2 is a drawing showing an example of frame
configuration.
In FIG.2, frame 21 and frame 31 are past drive sound
source signal sequences stored in the adaptive code book.
The Adaptive Sound Source Vector Generator (ASSVG) 13
searches for the frame pitch cycle between lower limit
32 and upper limit 267 of the pitch cycle search range.
As pitch cycle 22 retrieved from frame 21 here is
longer than the length of subframe 23, the Adaptive Sound
Source Vector Generator (ASSVG) 13 takes section 23
extracted from frame 21 for the frame lengthof the sub frame
as the adaptive sound source vector.
Also, as pitch cycle 32 retrieved from frame 31 is
shorter than the length of subframe 33, the Adaptive Sound
Source Vector Generator (ASSVG) 13 extracts the adaptive
sound source vector up to pitch cycle 32, and takes vector
section 34, obtained by iterating extracted vector
section 33 up to the length of the subframe length, as
the adaptive sound source vector.
Moreover, the Adaptive Sound Source Vector
Generator (ASSVG) 13 extracts from the Adaptive Code Book
12 (ACB) the adaptive sound source vector necessary when
finding the adaptive sound source vector corresponding
to a fractional-accuracy pitch cycle, and outputs this
to the Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 15.
Next, the Integral Pitch Cycle Searcher (IPCS) 14
calculates integral pitch cycle selection measure
DIST(T-int) from adaptive sound source vector p(t-int)
that has integral pitch cycle T-int, combining filter
impulse response matrix H, and target vector X.
Equation (1) is the equation for calculating
integral pitch cycle selection measure DIST(T-int).

When calculating integral pitch cycle selection
measure DIST (T-int) , matrix H' , obtained by multiplying
combining filter impulse response matrix H by auditory
weighting filter impulse response matrix W, may be used
in Equation (1) instead of combining filter impulse
response matrix H.
Here, the Integral Pitch Cycle Searcher (IPCS) 14
repeatedly executes integral pitch cycle selection
measure DIST(T-int) calculation processing using
Equation (1) for 236 variations of pitch cycle T-int from
pitch cycle 32 to 2 67 indicated by the Pitch Cycle Indicator
(PCI) 11.
The Integral Pitch Cycle Searcher (IPCS) 14 also
selects the DIST(T-int) with the largest value from the
236 calculated integral pitch cycle selection measures
DIST (T-int) , and outputs the selected DIST (T-int) to the
Distortion Comparator (DC) 17. Inaddition, the Integral
Pitch Cycle Searcher (IPCS) 14 outputs an index
corresponding to adaptive sound source vector pitch cycle
T-int, referenced when calculating DIST(T-int), to the
Distortion Comparator (DC) 17 as IDX(INT).
Next, the Fractional Pitch Cycle Adaptive Sound
Source Vector Generator (FPCASSVG) 15 finds adaptive
sound source vector p(T-frac) that has
fractional-accuracy pitch cycle T-frac (32+1/2,
33+1/2, ..., 51+1/2) by a product-sum operation on the
adaptive sound source vector received from the Adaptive
Sound Source Vector Generator (ASSVG) 13 and a SYNC
function, and outputs this p (T-frac) to the Fractional
Pitch Cycle Searcher (FPCS) 16.
The Fractional Pitch Cycle Searcher (FPCS) 16 then
calculates fractional pitch cycle selection measure
DIST(T-frac) from the adaptive sound source vector
p(T-frac) that has fractional pitch cycle T-frac,
combining filter impulse response matrix H, and target
vector X. Equation (2) is the equation for calculating
fractional pitch cycle selection measure DIST(T-frac).
When calculating fractional pitch cycle selection
measure DIST(T-frac) , matrix H' , obtained by multiplying
combining filter impulse response matrix H by auditory
weighting filter impulse response matrix W, may be used
in Equation (2) instead of combining filter impulse
response matrix H.
Here, the Fractional Pitch Cycle Searcher (FPCS)
16 repeatedly executes fractional pitch cycle selection
measure DIST(T-frac) calculation processing using
Equation (2) for 20 variations of fractional pitch cycle
T-frac from pitch cycle 32+1/2 to 51+1/2.
The Fractional Pitch Cycle Searcher (FPCS) 16 also
selects the DIST(T-frac) with the largest value from the
20 calculated fractional pi~ch cycle selection measures
DIST(T-frac), and outputs the selected DIST(T-frac) to
the Distortion Comparator (DC) 17.
In addition, the Fractional Pitch Cycle Searcher
(FPCS) 16 outputs an index corresponding to adaptive sound
source vector pitch cycle T-frac, referenced when
calculating DIST(T-frac), to the Distortion Comparator
(DC) 17 as IDX(FRAC).
Next, the Distortion Comparator (DC) 17 compares
the values of DIST(INT) received from the Integral Pitch
Cycle Searcher (IPCS) 14 and DIST(FRAC) received from
the Fractional Pitch Cycle Searcher (FPCS) 16. Then the
Distortion Comparator (DC) 17 determines the pitch cycle
when pitch cycle selection measure DIST with the larger
value of DIST(INT) and DIST(FRAC) is calculated as the
optimal pitch cycle, and outputs the index corresponding
to the optimal pitch cycle as optimal index IDX.
When, as in the above example, an integral-accuracy
pitch cycle search range from 32 to 267, and a
fractional-accuracy pitch cycle search range from 32 + 1/2
to 51 + 1/2, are selected as the pitch cycle search ranges,
a total of 256 (256 = 236 + 20) integral-accuracy and
fractional-accuracy pitch cycle search candidates are
provided, and optimal index IDX is coded as 8-bit binary
data .
The above-described "linear predictive residual
pitch cycle search apparatus using an adaptive code book"
is characterized by both performing a pitch cycle search
at integral accuracy and performing a 1/2
fractional-accuracy pitch cycle search in a section
corresponding to a shorter pitch cycle than the pitch
cycle search range at integral accuracy, and performing
selection of a final pitch cycle from the optimal pitch
cycle retrieved at integral accuracy and the optimal pitch
cycle retrieved at fractional accuracy.
Thus, with a conventional pitch search apparatus,
linear predictive residual pitch cycles can be
encoded/decoded efficiently for a female voice, which
contains many comparatively short pitch cycles. The
above characteristic and effect are disclosed in Document
2 (IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS,
pp.31-pp.41, VOL. 13, No. 1, JANUARY 1995), etc.
However, with a conventional pitch search apparatus,
the range for searching for a pitch cycle at fractional
accuracy is limited to short pitch cycles, and therefore,
for a male voice, which contains many comparatively long
pitch cycles, pitch cycles are searched for outside the
range in which pitch cycles are searched for at fractional
accuracy, and pitch cycles are searched for at integral
accuracy only, with a resultant problem that pitch cycle
resolution falls, and it is difficult to perform
encoding/decoding efficiently.
Disclosure of Invention
It is an object of the present invention to provide
a pitch search apparatus that enables speech signal pitch
cycles to be encoded/decoded efficiently.
This object is achieved by not fixing the range of
pitch cycles searched for at fractional accuracy, but
searching at fractional accuracy in the vicinity of a
pitch cycle retrieved in the previous subframe.
Brief Description of Drawings
FIG.l is a block diagram showing the configuration
of a conventional pitch cycle search apparatus;
FIG.2 is a drawing showing an example of frame
configuration;
FIG.3 is a block diagram showing the configuration
of a pitch cycle search apparatus according to Embodiment
1 of the present invention;
FIG.4 is a flowchart showing an example of the
operation of a pitch cycle search apparatus of this
embodiment;
FIG. 5 is a block diagram showing the configuration
of a decoding adaptive sound source vector generation
apparatus according to Embodiment 2 of the present
invention;
FIG.6 is a block diagram showing the internal
configuration of the speech decoding section 503 in FIG. 4;
FIG.7 is a block diagram showing the configuration
of a speech encoding apparatus 403; and
FIG.8 is a block diagram showing the internal
configuration of the speech decoding section 503 in FIG. 6.
Best Mode for Carrying out the Invention
With reference now to the accompanying drawings,
embodiments of the present invention will be explained
in detail below.
(Embodiment 1)
FIG.3 is a block diagram showing the configuration
of a pitch cycle search apparatus according to Embodiment
1 of the present invention. The pitch cycle search
apparatus 100 in FIG. 3 is mainly composed of a Pitch Cycle
Indicator (PCI) 101, Adaptive Code Book (ACB) 102,
Adaptive Sound Source Vector Generator (ASSVG) 103,
Integral Pitch Cycle Searcher (IPCS) 104, Fractional
Pitch Cycle Adaptive Sound Source Vector Generator
(FPCASSVG) 105, Fractional Pitch Cycle Searcher (FPCS)
106, Distortion Comparator (DC) 107, Last Sub Frame
Integral Pitch Cycle Storage (LSFIPCS) 108, Optimal Pitch
Cycle Accuracy Judge Section (OPCAJS) 109, and Comparison
Judge Section (CJS) 110.
The Pitch Cycle Indicator (PCI) 101 sequentially
indicates to the Adaptive Sound Source Vector Generator
(ASSVG) 103 pitch cycles T-int within a preset pitch cycle
search range. The Adaptive Code Book (ACB) 102 stores
drive sound source signals generated in the past.
The Adaptive Sound Source Vector Generator (ASSVG)
103 extracts from the Adaptive Code Book (ACB) 102 the
adaptive sound source vector p(t-int) that has
integral-accuracy pitch cycle T-int in accordance with
a directive received from the Pitch Cycle Indicator (PCI)
101, and outputs this adaptive sound source vector
p(t-int) to the Integral Pitch Cycle Searcher (IPCS) 104.
The Adaptive Sound Source Vector Generator (ASSVG)
103 reads integral-accuracy pitch cycle TO selected in
the previous subframe from the Last Sub Frame Integral
Pitch Cycle Storage (LSFIPCS) 108, sets preceding and
succeeding pitch cycles centered on this pitch cycle TO
as a range for searching for a fractional-accuracy pitch
frequency, extracts adaptive sound source vector
p (T-frac) that has fractional-accuracy pitch cycle T-frac
within this range from the Adaptive Code Book (ACB) 102,
and outputs the extracted adaptive sound source vector
to the Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 105.
The Integral Pitch Cycle Searcher (IPCS) 104
calculates integral pitch cycle selection measure
DIST(T-int) from adaptive sound source vector p(t-int)
received from the Adaptive Sound Source Vector Generator
(ASSVG) 103, combining filter impulse response matrix
H, and target vector x. The Integral Pitch Cycle Searcher
(IPCS) 104 then selects the DIST(T-int) with the largest
value from the integral pitch cycle selection measures
DIST(T-int) , and outputs the selected DIST(T-int) to the
Distortion Comparator (DC; 107.
The Fractional Pitch Cycle Adaptive Sound Source
Vector Generator (FPCASSVG) 105 finds adaptive sound
source vector p (T-frac) that has fractional-accuracy
pitch cycle T-frac (T-frac = T0-10+1/2, TO-9+1/2, ...,
TO + 9 + 1/2) by a product-sum operation on the adaptive sound
source vector received from the Adaptive Sound Source
Vector Generator (ASSVG) 103 and a SYNC function, and
outputs this p (T-frac) to the Fractional Pitch Cycle
Searcher (FPCS) 106.
The Fractional Pitch Cycle Searcher (FPCS) 106
calculates fractional pitch cycle selection measure
DIST (T-frac) from adaptive sound source vector p(T-frac)
received from the Fractional Pitch Cycle Adaptive Sound
Source Vector Generator (FPCASSVG) 105, combining filter
impulse response matrix H, and target vector x. The
Fractional Pitch Cycle Searcher (FPCS) 106 then selects
the DIST (T-frac) with the largest value from the
fractional pitch cycle selection measures DIST(T-frac),
and outputs the selected DIST(T-frac) to the Distortion
Comparator (DC) 107.
The Distortion Comparator (DC) 107 compares the
values of DIST (INT) received from the Integral Pitch Cycle
Searcher (IPCS) 104 and DIST(FRAC) received from the
Fractional Pitch Cycle Searcher (FPCS) 106. Then the
Distort ion Comparator (DC) 1 07 determines the pi tch cycle
when pitch cycle selection measure DIST with the larger
value of DIST(INT) and DIST(FRAC) is calculated as the
optimal pitch cycle, and outputs the index, of IDX(INT)
and IDX(FRAC), corresponding to the optimal pitch cycle
as optimal index IDX.
Then the Distortion Comparator (DC) 107 outputs
optimal pitch cycle integral component TO to the Last
Sub Frame Integral Pitch Cycle Storage (LSFIPCS) 108,
and outputs the optimal pitch cycle to the Optimal Pitch
Cycle Accuracy Judge Section (OPCAJS) 109.
The Last Sub Frame Integral Pitch Cycle Storage
(LSFIPCS) 108 stores integral component TO of the optimal
pitch cycle selected by the Distortion Comparator (DC)
107, and when apitch cycle of the next sub frame is searched
for, outputs this optimal pitch cycle integral component
TO to the Adaptive Sound Source Vector Generator (ASSVG)
103.
The Optimal Pitch Cycle Accuracy Judge Section
(OPCAJS) 109 judges whether the optimal pitch cycle is
of integral accuracy or fractional accuracy. The
Comparison Judge Section (CJS) 110 restricts the number
of times fractional-accuracy pitch information is
selected in an optimal pitch cycle.
Next, the operation of a pitch cycle search apparatus
10 0 according to this embodiment willbedescribed. FIG.4
is a flowchart showing an example of the operation of
a pitch cycle search apparatus of this embodiment.
In FIG.4, in step (hereinafter referred to as "ST")
201, the integral-accuracy pitch cycle TO selected in
the previous subframe is read from the Last Sub Frame
Integral Pitch Cycle Storage (LSFIPCS) 1 0 8 by the Adaptive
Sound Source Vector Generator (ASSVG) 103.
In ST202, an adaptive sound source vector is
generated by the Adaptive Sound Source Vector Generator
(ASSVG) 103. In ST203, optimal integral-accuracy pitch
cycle T-int is searched for by the Integral Pitch Cycle
Searcher (IPCS) 104.
In ST204, the Comparison Judge Section (CJS) 110
judges whether or not a fractional-accuracy pitch cycle
search is necessary. If a fractional-accuracy pitch
cycle search is necessary, the processing flow proceeds
to ST205. If a fractional-accuracy pitch cycle search'
is not necessary, the processing flow proceeds to ST207.
In ST205, an adaptive sound source vector that has
fractional-accuracy pitch cycle T-frac is generated by
the Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 105. In ST206, the optimal
fractional-accuracy pitch cycle T-frac is searched for
by the Fractional Pitch Cycle Searcher (FPCS) 106.
In ST207, the optimal pitch cycle is selected by
the Distortion Comparator (DC) 107 from optimal
integral-accuracy pitch cycle T-int and optimal
fractional-accuracy pitch cycle T-frac. In ST208,
integral component TO of the optimal pitch cycle selected
by the Distortion Comparator (DC) 107 is stored in the
Last Sub Frame Integral Pitch Cycle Storage (LSFIPCS)
108 .
In ST209, the Optimal Pitch Cycle Accuracy Judge
Section (OPCAJS) 109 judges whether the optimal pitch
cycle selected by the Distortion Comparator (DC) 107 is
an integral-accuracypitch cycle or a fractional-accuracy
pitch cycle.
In ST210, a counter indicating the number of times
a fractional-accuracy pitch cycle has been selected as
the optimal pitch cycle is reset to 0 by the Comparison
JudgeSection (CJS) 110. InST211, the counter indicating
the number of times a fractional-accuracy pitch cycle
has been selected as the optimal pitch cycle is incremented
by 1 by the Comparison Judge Section (CJS) 110.
In ST212, if pitch cycle search apparatus 100
processing has not finished, the processing flow returns
to ST201.
Detailed operations are described below for an
example in which a pitch cycle search apparatus 100 with
the above-described configuration has an 8-bit-sized
adaptive code book, and performs target pitch cycle
searching, in a CELP speech encoding/decoding apparatus
that performs encoding/decoding of a 16 kHz speech signal.
The Pitch Cycle Indicator (PCI) 101 sequentially
indicates to the Adaptive Sound Source Vector Generator
(ASSVG) 103 pitch cycles T-int within a preset pitch cycle
search range. For example, when the target vector pitch
cycle search range is preset from 32 to 267 at integral
accuracy, and from 32 + 1/2 to 51+1/2 at fractional accuracy
in a CELP speech encoding/decoding apparatus that
performs encoding and decoding of a speech signal with
a 16 kHz sampling frequency, the Pitch Cycle Indicator
(PCI) 101 outputs pitch cycles T-int (T-int = 32, 33, ...,
267) sequentially to the Adaptive Sound Source Vector
Generator (ASSVG) 103.
Next, the Adaptive Sound Source Vector Generator
(ASSVG) 103 extracts from the Adaptive Code Book (ACB)
102 the adaptive sound source vector p(t-int) that has
integral-accuracy pitch cycle T-int in accordance with
a directive received from the Pitch Cycle Indicator (PCI)
101, and outputs this adaptive sound source vector
p(t-int) to the Integral Pitch Cycle Searcher (IPCS) 104.
The Adaptive Sound Source Vector Generator (ASSVG)
103 reads integral-accuracy pitch cycle TO selected in
the previous subframe from the Last Sub Frame Integral
Pitch Cycle Storage (LSFIPCS) 108, sets preceding and
succeeding pitch cycles centered on this pitch cycle TO
as a range for searching for a fractional-accuracy pitch
frequency, extracts adaptive sound source vector
p(T-frac) that has fractiona 1-accuracy pitch cycle T-frac
within this range from the Adaptive Code Book (ACB) 102,
and outputs the extracted adaptive sound source vector
to the Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 105.
Specifically, the Adaptive Sound Source Vector
Generator (ASSVG) 103 set s 2 0 pi tch cycles T-frac cent ered
on integral component TO (T-frac = TO-10+1/2,
TO-9+1/2, . .., TO+9+1/2), and extracts adaptive sound
source vector p (T-frac) that has these pitch cycles from
the Adaptive Code Book (ACB) 102.
Then , using Equation (3) shown below, the Integral
Pitch Cycle Searcher (IPCS) 1 04 calculates integral pitch
cycle selection measure DIST(T-int) from adaptive sound
source vector p (t-int) received from the Adaptive Sound
Source Vector Generator (ASSVG) 103, combining filter
impulse response matrix H, and target vector x.
Here, the Integral Pitch Cycle Searcher (IPCS) 104
repeatedly executes integral pitch cycle selection
measure DIST(T-int) calculation processing using
Equation (3) for 236 variations of pitch cycle T-int from
pitch cycle 32 to 2 67 indicated by the Pitch Cycle Indicator
(PCI) 101.
The Integral Pitch Cycle Searcher (IPCS) 104 also
selects the DIST(T-int) with the largest value from the
236 calculated integral pitch cycle selection measures
DIST(T-int) , and outputs the selected DIST(T-int) to the
Distortion Comparator (DC) 107. In addition, the
Integral Pitch Cycle Searcher (IPCS) 104 outputs an index
corresponding to adaptive sound source vector pitch cycle
T-int, referenced when calculating DIST(T-int), to the
Distortion Comparator (DC) 107 as IDX(INT).
Next, the Fractional Pitch Cycle Adaptive Sound
Source Vector Generator (FPCASSVG) 105 finds adaptive
sound source vector p(T-frac) that has
fractional-accuracy pitch cycle T-frac (T-frac =
TO-10+1/2, TO-9+1/2, ..., TO+9+1/2) by a product-sum
operation on the adaptive scund source vector received
from the Adaptive Sound Source Vector Generator (ASSVG)
103 and a SYNC function, and outputs this p(T-frac) to
the Fractional Pitch Cycle Searcher (FPCS) 106.
The Fractional Pitch Cycle Searcher (FPCS) 106 then
calculates fractional pitch cycle selection measure
DIST(T-frac) from the adaptive sound source vector
p(T-frac) that has fractional pitch cycle T-frac,
combining filter impulse response matrix H, and target
vector X. Eguation (4) is the equation for calculating
fractional pitch cycle selection measure DIST (T-frac) .

Here, the Fractional Pitch Cycle Searcher (FPCS)
106 repeatedly executes fractional pitch cycle selection
measure DIST(T-frac) calculation processing using
Equation (4) for 20 variations of fractional pitch cycle
T-frac from pitch cycle T0-10+1/2 to TO+9+1/2.
The Fractional Pitch Cycle Searcher (FPCS) 106 then
selects the DIST(T-frac) with the largest value from the
20 calculated fractional pi-ch cycle selection measures
DIST (T-frac) , and outputs the selected DIST(T-frac) to
the Distortion Comparator (DC) 107. In addition, the
Fractional Pitch Cycle Searcher (FPCS) 106 outputs an
index corresponding to adaptive sound source vector pitch
cycle T-frac, referenced when calculating DIST(T-frac) ,
to the Distortion Comparator (DC) 107 as IDX(FRAC).
Next, the Distortion Comparator (DC) 107 compares
the values of DIST(INT) received from the Integral Pitch
Cycle Searcher (IPCS) 104 and DIST(FRAC) received from
the Fractional Pitch Cycle Searcher (FPCS) 106. Then the
Distortion Comparator (DC) 1 07 determines the pitch cycle
when pitch cycle selection measurement DIST with the
larger value of DIST(INT) and DIST(FRAC) is calculated
as the optimal pitch cycle, and outputs the index, of
IDX(INT) and IDX(FRAC), corresponding to the optimal
pitch cycle as optimal index IDX.
Then the Distortion Comparator (DC) 107 outputs
optimal pitch cycle integral component TO to the Last
Sub Frame Integral Pitch Cycle Storage (LSFIPCS) 108,
and outputs the optimal pitch cycle to the Optimal Pitch
Cycle Accuracy Judge Section (OPCAJS) 109.
When, as in the above example, an integral-accuracy
pitch cycle search range from 32 to 267, and a
fractional-accuracy pitch cycle search range from
TO-10+1/2 to T0+9+1/2, are selected as the pitch cycle
search ranges, a total of 256 (256 = 236 + 20)
integral-accuracy and fractional-accuracy pitch cycle
search candidates are provided, and optimal index IDX
is coded as 8-bit binary data.
The Last Sub Frame Integral Pitch Cycle Storage
(LSFIPCS) 108 stores integral component TO of the optimal
pitch cycle selected by the Distortion Comparator (DC)
107, and when a pitch cycle of the next sub frame is searched
for, outputs this optimal pitch cycle integral component
TO to the Adaptive Sound Source Vector Generator (ASSVG)
103.
The Optimal Pitch Cycle Accuracy Judge Section
(OPCAJS) 109 judges whether the optimal pitch cycle is
of integral accuracy or fractional accuracy. When the
optimal pitch cycle is of integral accuracy, the Optimal
Pitch Cycle Accuracy Judge Section (OPCAJS) 109 resets
the Comparison Judge Section (CJS) 11 0 counter to 0 . When
the optimal pitch cycle is of fractional accuracy, the
Optimal Pitch Cycle Accuracy Judge Section (OPCAJS) 109
adds 1 to the Comparison Judge Section (CJS) 110 counter.
Specifically, the Comparison Judge Section (CJS)
110 is provided with a counter that indicates the number
of times a fractional-accuracy pitch cycle has been
selected as the optimal pitch cycle, and compares the
counter value with a preset non-negative integer N. If
the counter value is greater than integer N, the Comparison
Judge Section (CJS) 110 outputs a directive to the
Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 105 indicating that a
fractional-accuracy pitch cycle is not to be performed.
If the counter value is less than or equal to integer
N, the Comparison Judge Section (CJS) 110 outputs a
directive to the Fractional Pitch Cycle Adaptive Sound
Source Vector Generator (FPCASSVG) 105 indicating that
a fractional-accuracy pitch cycle is to be performed.
Thus, according to a pitch cycle search apparatus
of this embodiment, by not fixing the range of pitch cycles
searched for at fractional accuracy, but searching at
fractional accuracy in the vicinity of a pitch cycle
retrieved in the previous subframe, it is possible for
pitch cycle searching to be carried out with high
resolution even for speech signals with long pitch cycles
or for speech signal linear predictive residuals.
Also, according to a pitch cycle search apparatus
of this embodiment, by searching at fractional accuracy
in the vicinity of a pitch cycle retrieved in the previous
subframe, it is possible to improve search accuracy for
speech signal linear predictive residuals, despite the
shortness of pitch cycles, and to perform high-quality
speech encoding and decoding.
In the above description, an example has been
described in which a linear predictive residual pitch
cycle is searched for using an adaptive code book, but
the object of a pitch cycle search is not limited to a
linear predictive residual, and this embodiment can be
applied to any speech signal information that has a pitch
cycle.
Furthermore, in the above description, when
calculating a pitch cycle selection measure, an
integral-accuracy pitch cycle search and
fractional-accuracy pitch cycle search have been
described using a closed-loop search procedure, but this
is not a limitation, and similar results can be achieved
with any procedure in which an integral-accuracy pitch
cycle search and fractional-accuracy pitch cycle search
are performed, and the integral-accuracy pitch cycle and
fractional-accuracy pitch cycle are compared.
For example, if a two-stage (open-loop and
closed-loop) pitch cycle search is carried out using the
above-described configuration, a Distortion Comparator
(DC) 107 that includes the Integral Pitch Cycle Searcher
(IPCS) 104 and Fractional Pitch Cycle Searcher (FPCS)
106 is configured, an adaptive sound source vector that
has an integral-accuracy pitch cycle received from the
Adaptive Sound Source Vector Generator (ASSVG) 103 and
an adaptive sound source vector that has a
fractional-accuracy pitch cycle received from the
Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 105 are used, and indexing
corresponding to the optimal pitch cycle of the subframe
to be processed is performed by me an sofa procedure divided
into two stages, an open-Loop search and closed-loop
search, in the Distortion Comparator (DC) 107.
Moreover, in the above description, the pitch cycle
search range has been taken to be 32 to 267, but there
is no particular limitation on the pitch cycle search
range, and similar results to those in the above
description can be obtained as long as the
fractional-accuracy pitch cycle search range is not
fixed.
Also, in the above description, the
fractional-accuracy pitch cycle search range has been
taken as 20 pitch cycles T-frac centered on
integral-accuracy pitch cycle TO (T-frac = TO-10+1/2,
TO-9+1/2, ..., TO+9+1/2), but there is no particular
limitation on the pitch cycle range, and any range set
based on the integral-accuracy pitch cycle may be used.
Furthermore, a description has been given in which
the maximum number of times the optimal pitch cycle is
selected with fractional-accuracy is a fixed value N,
but this value N may also be increased or decreased
adaptively according to the communication environment.
Moreover, in the above description, the number of
times a fractional-accuracy pitch cycle is selected is
limited to N consecutive times, but it is also possible
for N to be set to infinitude, and for the number of times
a fractional-accuracy pitch cycle is selected to be made
infinite. In particular, if it is not necessary to
consider the occurrence of an error when transmitting
a pitch cycle index—for example, when coding information
including this pitch cycle index is written to a storage
medium—the results of a pitch cycle search can be encoded
with high resolution, without a limit on the number of
fractional-accuracy pitch cycle selections, by making
the number of times a fractional-accuracy pitch cycle
is selected infinite.
Furthermore, in the above description, an example
has been described in which a pitch cycle search is not
performed at fractional accuracy when the number of times
a fractional-accuracy pitch cycle is selected exceeds
a predetermined limit, but this is not a limitation, and
a fractional-accuracy pitch cycle search may also be
carried out in a predetermined range—for example, from
32+1/2 to 51+1/2—when the number of times a
fractional-accuracy pitch cycle is selected exceeds the
predetermined limit.
By performing a fractional-accuracy pitch cycle
search when the number of times a fractional-accuracy
pitch cycle is selected exceeds a predetermined limit
in this way, it is possible to encode the results of a
pitch cycle search with high resolution even if an error
occurs when a pitch cycle index is transmitted.
In the above description, when calculating
integral pitch cycle selection measure DIST(T-int) or
DIST (T-frac), matrix H', obtained by multiplying
combining filter impulse response matrix H by auditory
weighting filter impulse response matrix W, may be used
instead of combining filter impulse response matrix H.
(Embodiment 2)
FIG. 5 is a block diagram showing the configuration
of a decoding adaptive sound source vector generation
apparatus according to Embodiment 2 of the present
invention.
The decoding adaptive sound source vector
generation apparatus 300 in FIG.5 is mainly composed of
an Adaptive Code Book 301 (ACB) , Last Sub Frame Integral
Pitch Cycle Storage (LSFIPCS) 302, Pitch Cycle Judge
Section (PCJS) 303, Adaptive Sound Source Vector
Generator (ASSVG) 304, and Fractional Pitch Cycle
Adaptive Sound Source Vector Generator (FPCASSVG) 305.
The Adaptive Code Book 301 (ACB) stores drive sound
source signals generated in the past.
The Last Sub Frame Integral Pitch Cycle Storage
(LSFIPCS) 302 receives integral component TO of a pitch
cycle judged by the Pitch Cycle Judge Section (PCJS) 303,
stores this TO, and when the next subframe is processed,
outputs this TO to the Pitch Cycle Judge Section (PCJS)
303.
The Pitch Cycle Judge Section (PCJS) 303 judges
whether a pitch cycle corresponding to index IDX is of
integral accuracy or fractional accuracy. The Pitch
Cycle Judge Section (PCJS) 303 then sets the pitch cycle
using index IDX transmitted from the encoding side and
integral component TO of the pitch cycle selected in the
previous subframe.
If, for example, received index IDX indicates an
integral-accuracy pitch cycle, the Pitch Cycle Judge
Section (PCJS) 303 conveys the pitch cycle corresponding
to index IDX to the Adaptive Sound Source Vector Generator
(ASSVG) 304.
If received index IDX indicates a
fractional-accuracy pitch cycle, the Pitch Cycle Judge
Section (PCJS) 303 finds the pitch cycle from information
on the pitch cycle corresponding to index IDX and pitch
cycle integral component TO for the previous subframe,
and conveys the obtained pitch cycle to the Adaptive Sound
Source Vector Generator (ASSVG) 304. Specifically, the
Pitch Cycle Judge Section (PCJS) 303 finds a value
corresponding to index IDX from the fractional-accuracy
pitch cycle range (-10+1/2, -9 + 1/2, ..., 9 + 1/2), and takes
the result of adding TO to this value as the
fractional-accuracy pitch cycle.
The Pitch Cycle Judge Section (PCJS) 303 is also
provided with a counter that counts the number of times
the pitch cycle corresponding to index IDX is a
fractional-accuracy pitch cycle.
When, for example, the pitch cycle corresponding
to index IDX is of fractional accuracy, the Pitch Cycle
Judge Section (PCJS) 303 adds 1 to the counter. When the
pitch cycle corresponding to index IDX is of integral
accuracy, the Pitch Cycle Judge Section (PCJS) 303 resets
the counter to 0.
When the pitch cycle is of integral accuracy, the
Adaptive Sound Source Vector Generator (ASSVG) 304
extracts from the Adaptive Code Book 301 (ACB) the adaptive
sound source vector p(T-int) that has pitch cycle T-int
in accordance with a directive received from the Pitch
Cycle Judge Section (PCJS) 303, and outputs adaptive sound
source vector p(T-int).
When the pitch cycle is of fractional accuracy, the
Adaptive Sound Source Vector Generator (ASSVG) 304 takes
from the Adaptive Code Book 3 01 (ACB) the adaptive sound
source vector necessary when extracting adaptive sound
source vector p(T-frac) that has pitch cycle T-frac in
accordance with a directive received from the Pitch Cycle
Judge Section (PCJS) 303, and outputs this to the
Fractional Pitch Cycle Adaptive Sound Source Vector
Generator (FPCASSVG) 305.
The Fractional Pitch Cycle Adaptive Sound Source
Vector Generator (FPCASSVG) 305 finds adaptive sound
source vector p(T-frac) that has fractional-accuracy
pitch cycle T-frac by a product-sum operation on the
adaptive sound source vector received from the Adaptive
Sound Source Vector Generator (ASSVG) 304 and a SYNC
function, and outputs this as the decoding adaptive sound
source vector.
(Embodiment 3)
In Embodiment 3, an example is described in which
a pitch cycle search apparatus according to Embodiment
1 or a decoding adaptive sound source vector generation
apparatus according to Embodiment 2 is used for
communications installed in a transmitting apparatus and
receiving apparatus.
FIG.6 is a block diagram showing the internal
configuration of a speech signal transmitting apparatus
and receiving apparatus according to Embodiment 3 of the
present invention.
The speech signal transmitting apparatus 400 in
FIG.6 is mainly composed of an input section 401, A/D
converter 402, speech encoding apparatus 403, RF
modulator 404, and transmitting antenna 405. The speech
signal receiving apparatus 500 inFIG.6 is mainly composed
of a receiving antenna 501, RF demodulator 502, speech
decoding section 503, D/A converter 504, and output
section 505.
In FIG.6, a speech signal is converted to an
electrical signal by the input section 401, and is then
output to the A/D converter 402. The A/D converter 402
converts the (analog) signal output from the input section
401 to a digital signal, and outputs this signal to the
speech encoding apparatus 403. The speech encoding
apparatus 403 is provided with a signal processing
apparatus according to either of the above-described
embodiments, encodes the digital speech signal output
from the A/D converter 402 using a speech encoding method
described later herein, and outputs encoded information
to the RF modulator 404 . The RF modulator 404 places the
speech encoded information output from the speech
encoding apparatus 403 on a propagation medium such as
a radio wave, converts the signal for sending, and outputs
it to the transmitting antenna 405. The transmitting
antenna 405 sends the output signal output from the RF
modulator 404 as a radio wave (RF signal) .
The RF signal is received by the receiving antenna
501 and output to the RF denodulator 502. The RF signal
in the drawing is an RF signal as seen from the receiving
side, and, if there is no signal attenuation or noise
superimposition in the propagation path, is exactly the
same as the transmitted RF signal. The RF demodulator
502 demodulates speech encoded information from the RF
signal output from the receiving antenna 501, and outputs
this information to the speech decoding section 503. The
speech decoding section 503 is provided with a signal
processing apparatus according to either of the
above-described embodiments , decodes a speech signal from
the speech encoded information output from the RF
demodulator 502 using a speech decoding method described
later herein, and outputs the resulting signal to the
D/A converter 504. The D/A converter 504 converts the
digital speech signal output from the speech decoding
section 503 to an analog electrical signal, and outputs
this signal to the output section 505. The output section
505 converts the electrical signal to vibrations of the
air, and outputs soundwaves that are audible to the human
ear.
By providing at least one of the above-described
kinds of speech signal transmitting apparatus and
receiving apparatus, it is possible to configure a base
station apparatus and mobile terminal apparatus in a
mobile communication system.
The special characteristic of speech signal
transmitting apparatus 400 lies in the speech encoding
apparatus 403. FIG.7 is a block diagram showing the
configuration of the speech encoding apparatus 403.
The speech encoding apparatus 403 in FIG. 7 is mainly
composed of a preprocessing section 601, LPC analysis
section 602, LPC quantization section 603, combining
filter 604, adder 605, adaptive sound source code book
606, quantization gain generator 607, fixed sound source
code book 608, multiplier 609, multiplier 610, adder 611,
auditory weighting section 612, parameter determination
section 613, and multiplexer 614.
In FIG.7, an input speech signal output from the
A/D converter 402 in FIG.6 is input to the preprocessing
section 601. The preprocessing section 601 performs
high-pass filter processing that eliminates the DC
component in the input speech signal, or waveform shaping
processing and pre-emphasis processing concerned with
improving the performance of later encoding processing,
and outputs the processed speech signal (Xin) to the LPC
analysis section 602, adder 605, and parameter
determination section 613. CELP encoding that uses this
preprocessing is disclosed in Unexamined Japanese Patent
Publication No.6-214600.
The LPC analysis section 602 performs linear
predictive analysis using Xin, and outputs the result
of the analysis (linear predictive coefficient) to the
LPC quantization section 603.
The LPC quantization section 603 converts the LPC
coefficient output from the LPC analysis section 602 to
an LSF parameter. The LSF parameter obtained by this
conversion is subjected tc vector quantization as a
quantization target vector, and an LPC code (L) obtained
by vector quantization is output to the multiplexer 614.
Also, the LPC quantization section 603 obtains an
LSF area decoding spectral envelope parameter, converts
the obtained decoding spectral envelope parameter to a
decoding LPC coefficient, and outputs the decoding LPC
coefficient obtained by the aforementioned conversion
to the combining filter 604.
The combining filter 604 performs filter
combination using the aforementioned encoding LPC
coefficient and a drive sound source output from the adder
611, and outputs the composite signal to adder 605.
Adder 605 calculates an error signal for
aforementioned Xin and the aforementioned composite
signal, and outputs this error signal to the auditory
weighting section 612. The auditory weighting section
612 performs auditory weighting on the error signal output
from adder 605, calculates distortion between Xin and
the composite signal in the auditory weighting area, and
outputs this distortion to the parameter determination
section 613.
The parameter determination section 613 determines
the signals generated in the adaptive sound source code
book 606, fixed sound source code book 608, and
quantization gain generator 607 so that the encoding
distortion output from the auditory weighting section
612 is minimized. Encoding performance can be further
improved by determining the signals that should be output
from the aforementioned three sections not only by
minimizing the encoding distortion output from the
auditory weighting section 612, but also by combined use
with separate encoding distortion using Xin.
The adaptive sound source code book 60 6 buffers sound
source signals output by adder 611 in the past, extracts
an adaptive sound source vector from a location specified
by a signal (A) output from the parameter determination
section 613, and outputs this vector to multiplier 609.
The fixed sound source code book 608 outputs to
multiplier 610 a vector of the form specified by a signal
(F) output from the parameter determination section 613.
The quantization gain generator 607 outputs to
multiplier 609 and multiplier 610, respectively, the
adaptive sound source gain and fixed sound source gain
specified by a signal (G) output from the parameter
determination section 613.
Multiplier 6 09 multipl ies the quantization adaptive
sound source gain output from the quantization gain
generator 607 by the adaptive sound source vector output
from the adaptive sound source code book 606, and outputs
the result of the multiplication to adder 611. Multiplier
610 multiplies the quantization fixed sound source gain
output from the quantization gain generator 607 by the
fixed sound source vector output from the fixed sound
source code book 608, and outputs the result of the
multiplication to adder 611.
Adder 611 has as inputs the adaptive sound source
vector following gainmultiplication frommultiplier 609,
and the fixed sound source vector from multiplier 610,
and performs vector addition of the adaptive sound source
vector and fixed sound source vector. Adder 611 then
outputs the result of the vector addition to the combining
filter 604 and adaptive sound source code book 606.
Finally, the multiplexer 614 has as inputs code L
indicating the quantization LPC from the LPC quantization
section 603, together with code A indicating the adaptive
sound source vector, code F indicating the fixed sound
source vector, and code G indicating the quantization
gain, from the parameter determination section 613,
quantizes these various items of information, and outputs
them to the propagation path as encoded information.
Next, the speech decoding section 503 will be
described in detail. FIG.8 is a block diagram showing
the internal configuration of the speech decoding section
503 in FIG.6.
In FIG.8, encoded information output from the RF
demodulator 502 is input to a multiplexing separator 701,
where multiplexed encoded information is separated into
individual kinds of code information.
Separated LPC code L is output to an LPC decoder
702, separated adaptive sound source vector code A is
output to an adaptive sound source code book 7 0 5, separated
sound source gain code G is output to a quantization gain
generator 706, and separated fixed sound source vector
code F is output to a fixed sound source code book 707.
The LPC decoder 702 obtains a decoding spectral
envelope parameter from code L output from the
multiplexing separator 701 by means of the vector
quantization decoding processing shown in Embodiment 1,
and converts the obtained decoding spectral envelope
parameter to a decoding LPC coefficient. The LPC decoder
702 then outputs the decoding LPC coefficient obtained
by this conversion to a combining filter 703.
The adaptive sound source code book 705 extracts
an adaptive sound source vector from the location
specified by code A output from the multiplexing separator
701, and outputs it to a multiplier 708 . The fixed sound
source code book 707 generates the fixed sound source
vector specified by code F output from the multiplexing
separator 701, and outputs it to a multiplier 709.
The quantization gain generator 706 decodes the
adaptive sound source vector gain and fixed sound source
vector gain specified by sound source gain code G output
from the multiplexing separator 701, and outputs these
to multiplier 708 and multiplier 709, respectively.
Multiplier 708 multiplies the aforementioned
adaptive code vector by the aforementioned adaptive code
vector gain, and outputs the result to an adder 710.
Multiplier 709 multiplies the aforementioned fixed code
vector by the aforementioned fixed code vector gain, and
outputs the result to the adder 710.
The adder 710 performs addition of the adaptive sound
source vector and fixed sound source vector after gain
multiplication output from mult ipl ier 708 and mult iplier
709, and outputs the result ~o the combining filter 703.
The combining filter 703 performs filter
combination using the combining filter, with the encoding
LPC coefficient supplied from the LPC decoder 702 as the
filter coefficient, and with the sound source vector
output from adder 710 as a drive signal, and outputs the
combined signal to a postprocessing section 704.
The postprocessing section 704 executes processing
to improve the subjective quality of speech, such as
formant emphasis and pitch emphasis, processing to
improve the subjective quality of stationary noise, and
so forth, and then outputs a final decoded speech signal.
The present invention is not limited to the
above-described embodiments, and various variations and
modifications may be possible without departing from the
scope of the present invention . For example, in the above
embodiments a case has been described in which the present
invention operates as a signal processing apparatus, but
this is not a limitation, and it is also possible for
this signal processing method to be implemented as
software.
For example, a program that executes the |
above-described signal processing method may be stored i
i
beforehand in ROM (Read Only Memory), and operated by ;
a CPU (Central Processing Unit) .
It is also possible for a program that executes the i
above-described signal processing method to be stored
on a computer-readable storage medium, for the program
stored on the storage medium to be recorded in the RAM
(Random Access Memory) of a computer, and for the computer
to be operated in accordance with that program.
As is clear from the above descriptions, according ,
to a pitch cycle search apparatus of the present invention, '
i
by not fixing the range of pitch cycles searched for at
fractional accuracy, but searching at fractional accuracy
in the vicinity of a pitch cycle retrieved in the previous
subframe, it is possible to improve search accuracy for
speech signal linear predictive residuals, despite the
shortness of pitch cycles, and to perform high-quality
speech encoding and decoding.
This application is based on Japanese Patent
Application No.2001-234559 filed on August 2, 2001,
entire contents of which are expressly incorporated by
reference herein.
Industrial Applicability
The present invention is suitable for use in a mobile
communication system in which speech signals are encoded
and transmitted.
We Claim:
1. A pitch cycle search range setting apparatus comprising:
a pitch cycle indicating section (101) that is adapted to search for a pitch
cycle included in a linear predictive residual on a subframe basis, and to
sequentially indicate pitch cycle candidates within a preset pitch cycle
search range with integral accuracy to an adaptive sound source vector
generating section (103); and
an adaptive sound source vector generating section (103) that is adapted
extract an adaptive sound source vector that has a pitch cycle indicated
by said pitch cycle indicating section (101) from an adaptive code book
that stores past drive sound sources;
characterized by
a last subframe integral pitch cycle storage section (108) that is adapted
to store the integral component of a pitch cycle finally selected in pitch
cycle search processing of a previous subframe;
wherein said pitch cycle search range setting apparatus is further adapted
to set as a pitch cycle search object both of an integral-accuracy pitch
cycle candidate indicated by said pitch cycle indicating section (101) and a
fractional-accuracy pitch cycle search candidate that covers with
fractional-accuracy a pitch cycle in the vicinity of an integral-accuracy
pitch cycle read from said last subframe integral pitch cycle storage
section (108).
2. The pitch cycle search range setting apparatus as claimed in claim 1,
comprising:
a comparison judging section (110) that is adapted to perform relative
size comparison of a value of a counter provided internally and a non-
negative integer N; and
an optimal pitch cycle accuracy judging section (109) that is adapted to
judge whether a pitch cycle selected as an optimal pitch cycle in
processing subframe pitch cycle search processing is of integral accuracy
or of fractional accuracy, and to manipulate a value of said counter
provided in said comparison judging section (110) in accordance with a
result of that judgment;
wherein said pitch cycle search range setting apparatus is further adapted
to perform pitch cycle search processing only for said integral-accuracy
pitch cycle search candidate when a value of said internal counter of said
comparison judging section (110) is greater than said N, and to perform a
pitch cycle search for both said integral-accuracy pitch cycle search
candidate and said fractional-accuracy pitch cycle search candidate when
a value of said internal counter of said comparison judging section (110)
is less than or equal to said N.
3. The pitch cycle search range setting apparatus as claimed in claim 2, wherein
said optimal pitch cycle accuracy judging section (109) is adapted to execute an
operation that resets a value of said internal counter of said comparison judging
section (110) to 0 when accuracy of a pitch cycle selected finally in processing
subframe section pitch cycle search processing is integral accuracy, and to
execute an operation that increments said internal counter of said comparison
judging section (110) when accuracy of a pitch cycle selected finally in
processing subframe section pitch cycle search processing is fractional accuracy.
4. A pitch cycle search apparatus comprising:
an adaptive sound source vector generating section (103) that is adapted
to extract from an adaptive code book an adaptive sound source vector
that has an integral-accuracy pitch cycle indicated by a pitch cycle
indicating section (101), and to output the extracted adaptive sound
source vector to an integral-accuracy pitch cycle search section (104) and
a fractional pitch cycle adaptive sound source vector generating section
(105);
an integral-accuracy pitch cycle search section (104) that is adapted to
perform a closed-loop search for an integral-accuracy pitch cycle using an
adaptive sound source vector received from said adaptive sound source
vector generating section (103), and to output an integral-accuracy
optimal pitch cycle index and selection measure to a distortion comparison
section ;
a fractional pitch cycle adaptive sound source vector generating section
(105) that is adapted to complement an integral-accuracy adaptive sound
source vector received from said adaptive sound source vector generating
section (103), and to generate an adaptive sound source vector that has a
fractional-accuracy pitch cycle, and to output the generated adaptive
sound source vector that has a fractional-accuracy pitch cycle to a
fractional-accuracy pitch cycle search section (106);
a fractional-accuracy pitch cycle search section (106) that is adapted to
perform a closed-loop search for a fractional-accuracy pitch cycle using an
adaptive sound source vector that has a fractional-accuracy pitch cycle
received from said fractional pitch cycle adaptive sound source vector
generating section (105), and to output a fractional-accuracy optimal pitch
cycle index and selection measure to said distortion comparison section ;
and
a comparison section (107) that is adapted to compare a selection
measure received from said integral-accuracy pitch cycle search section
(104) with a selection measure received from said fractional-accuracy
pitch cycle search section (106), and to output an index with a larger
selection measure as an index indicating a processing subframe section
optimal pitch cycle, and to output an integral component of a pitch cycle
with a larger selection measure to a last subframe integral pitch cycle
storage section (108);
wherein said pitch cycle search apparatus is further adapted to search for
a pitch cycle possessed by a processing subframe section linear predictive
residual from among pitch cycle candidates within a range set by the pitch
cycle search range setting apparatus as claimed in claim 1.
5. The pitch cycle search apparatus as claimed in claim 4, comprising:
a distortion comparison section (107) that is adapted to find an index
indicating an optimal pitch cycle among processing subframe section
linear predictive residuals by means of a two-stage search, comprising an
open-loop search and closed-loop search, on an adaptive sound source
vector that has an integral-accuracy pitch cycle generated by said
adaptive sound source vector generating section (103) and an adaptive
sound source vector that has a fractional-accuracy pitch cycle obtained by
interpolating an adaptive sound source vector that has an integral -
accuracy pitch cycle generated by said adaptive sound source vector
generating section (103), and to output an optimal pitch cycle integral
component to a last subframe integral pitch cycle storage section (108);
wherein said pitch cycle search apparatus is further adapted to search for
an optimal pitch cycle from within a pitch cycle search range set by the
pitch cycle search range setting apparatus as claimed in claim 1.
6. The pitch cycle search apparatus as claimed in claim 4, comprising:
a comparison judging section (110) that is adapted to perform relative
size comparison of a value of a counter provided internally and a non-
negative integer N;
an optimal pitch cycle accuracy judging section (109) that is adapted to ...
judge whether a pitch cycle selected as an optimal pitch cycle in
processing subframe pitch cycle search processing is of integral accuracy
or of fractional accuracy, and to manipulate a value of said counter
provided in said comparison judging section (110) in accordance with a
result of that judgment; and
and a pitch cycle search range setting apparatus that is adapted to
perform pitch cycle search processing only for said integral-accuracy pitch
cycle search candidate when a value of said internal counter of said
comparison judging section (110) is greater than said N, and to perform a
pitch cycle search for both said integral-accuracy pitch cycle search
candidate and said fractional-accuracy pitch cycle search candidate when
a value of said internal counter of said comparison judging section (110)
is less than or equal to said N.
7. The pitch cycle search apparatus as claimed in claim 6, wherein an
arbitrary natural number is set for said non-negative integer N which should set
an upper limit of a consecutive number of subframes for which accuracy of a
pitch cycle selected finally is fractional accuracy.
8. The pitch cycle search apparatus as claimed in claim 4, wherein a value of
non-negative integer N can be varied in accordance with a degree of index
transmission error occurrence frequency.
9. The pitch cycle search apparatus as claimed in claim 7, wherein; in a
fractional-accuracy pitch cycle search, when a value of said counter is less than
or equal to non-negative integer N that is an object of comparison, a fractional-
accuracy pitch cycle search is performed within a predetermined range, and
when a value of said counter is greater than non-negative integer N that is an
object of comparison, said counter is reset to 0 irrespective of whether a pitch
cycle chosen as an optimal pitch cycle is of integral accuracy or of fractional
accuracy.
10. The pitch cycle search apparatus as claimed in claim 9, wherein, in said
fractional-accuracy pitch cycle search section (106), when a value of said counter
is greater than a value of non-negative integer N that is an object of comparison,
said fractional-accuracy pitch cycle search and a fractional-accuracy pitch cycle
search in a section with a short pitch cycle are performed.
11. A decoding adaptive sound source vector generating apparatus comprising:
a pitch cycle judging section that is adapted to find an optimal adaptive
sound source vector pitch cycle and to pass the optimal adaptive sound
source vector pitch cycle to an adaptive sound source vector generating
section (103);
an adaptive sound source vector generating section (103) that is adapted
to extract from an adaptive code book an adaptive sound source vectbr
that has a pitch cycle received from said pitch cycle judging section, to
output the extracted adaptive sound source vector if a pitch cycle is of
integral accuracy, and to output the extracted adaptive sound source
vector to a fractional pitch cycle adaptive sound source vector generating
section (105) if a pitch cycle is of fractional accuracy; and
a fractional pitch cycle adaptive sound source vector generating section
(105) that is adapted to generate and output an adaptive sound source
vector that has a fractional-accuracy pitch cycle from an adaptive sound
source vector received from said adaptive sound source vector generating
section (103).
characterized by
a last subframe integral pitch cycle storage section (108) that is adapted to store
a pitch cycle selected in a previous subframe section;
wherein said fractional pitch cycle adaptive sound source vector generating
section is adapted to find the optimal adaptive sound source vector pitch cycle
using a pitch cycle selected in a previous subframe received from said last
subframe integral pitch cycle storage section (108) and an index received as
input.
12. A speech encoding apparatus comprising:
the pitch cycle search apparatus as claimed in claim 4;
a fixed sound source vector generating section that is adapted to generate
a fixed sound source vector from a fixed code book;
a section that is adapted to perform quantization and encoding of a
parameter indicating a spectral characteristic of an input speech signal;
a section that is adapted to synthesize a composite speech signal using a
sound source vector generated from said fixed sound source vector
generating section and said adaptive sound source vector pitch cycle
search apparatus and said parameter; and
a section that is adapted to determine output from said fixed sound
source vector generating section and said adaptive sound source vector
pitch cycle search apparatus so that distortion of said input speech signal
and said composite speech signal becomes small.
13. A speech decoding apparatus comprising:
a section that is adapted to decode an index indicating an adaptive sound
source vector pitch cycle encoded by a speech encoding apparatus
comprising the decoding adaptive sound source vector generating
apparatus as claimed in claim 11;
a fixed sound source vector generating section that is adapted to generate/
a fixed sound source vector from a fixed code book ;
a second section that is adapted to decode a parameter indicating a
spectral characteristic encoded by said speech encoding apparatus ; and
a third section that is adapted to decode a sound source vector
determined in said speech encoding apparatus from said fixed sound
source vector generating section and said decoding adaptive sound source
vector generating apparatus , and to synthesize a composite speech signal
from a decoded sound source vector and said parameter.
14. A speech signal transmitting apparatus comprising:
a speech input apparatus that is adapted to convert a speech signal to an
electrical signal;
an A/D conversion apparatus that is adapted to convert a signal output by
said speech input apparatus to a digital signal;
the speech encoding apparatus as claimed in claim 12 that is adapted to
perform encoding of a digital signal output from said A/D conversion
apparatus;
an RF modulation apparatus that is adapted to perform modulation
processing and so forth on encoded information output from said, speech
encoding apparatus; and
a transmitting antenna that is adapted to convert a signal output from
said RF modulation apparatus to a radio wave and transmits that radio
wave.
15. A speech signal receiving apparatus comprising:
a receiving antenna that is adapted to receive a reception radio wave;
an RF demodulation apparatus that is adapted to perform demodulation
processing on a signal received by said receiving antenna ;
the speech decoding apparatus as claimed in claim 13 that is adapted to
perform decoding processing on information obtained by said RF
demodulation apparatus;
a D/A conversion apparatus that is adapted to perform D/A conversion of
a digital speech signal decoded by said speech decoding apparatus ; and
a speech output apparatus that is adapted to convert an electrical signal
output from said D/A conversion apparatus to a speech signal.
16. A mobile station apparatus that has a speech signal transmitting apparatus
and performs radio communications with a base station apparatus, said speech
signal transmitting apparatus comprising:
a speech input apparatus that is adapted to convert a speech signal to an
electrical signal;
an A/D conversion apparatus that is adapted to convert a signal output by
said speech input apparatus to a digital signal;
the speech encoding apparatus as claimed in claim 12 that is adapted to
perform encoding of a digital signal output from said A/D conversion
apparatus;
an RF modulation apparatus that is adapted to perform modulation
processing and so forth on encoded information output from said speech
encoding apparatus; and
a transmitting antenna that is adapted to convert a signal output from
said RF modulation apparatus to a radio wave and transmits that radio
wave.
17. A mobile station apparatus that has a speech signal receiving apparatus and
performs radio communications with a base station apparatus, said speech signal
receiving apparatus comprising:
a receiving antenna that is adapted to receive a reception radio wave;
an RF demodulation apparatus that is adapted to perform demodulation
processing on a signal received by said receiving antenna;
the speech decoding apparatus as claimed in claim 13 that is adapted to
perform decoding processing on information obtained by said RF
demodulation apparatus;
a D/A conversion apparatus that is adapted to perform D/A conversion of
a digital speech signal decoded by said speech decoding apparatus ; and
a speech output apparatus that is adapted to convert an electrical signal
output from said D/A conversion apparatus to a speech signal.
18. A base station apparatus that has a speech signal transmitting apparatus and
performs radio communications with a mobile station apparatus, said speech
signal transmitting apparatus comprising:
a speech input apparatus that is adapted to convert a speech signal to an
electrical signal;
an A/D conversion apparatus that is adapted to convert a signal output by
said speech input apparatus to a digital signal;
the speech encoding apparatus as claimed in claim 12 that is adapted to
perform encoding of a digital signal output from said A/D conversion
apparatus;
an RF modulation apparatus that is adapted to perform modulation
processing and so forth on encoded information output from said speech
encoding apparatus; and
a transmitting antenna that is adapted to convert a signal output from said
RF modulation apparatus to a radio wave and transmits that radio wave.
19. A base station apparatus that has a speech signal receiving apparatus and
performs radio communications with a mobile station apparatus, said speech
signal receiving apparatus comprising:
a receiving antenna that is adapted to receive a reception radio wave;
an RF demodulation apparatus that is adapted to perform demodulation
processing on a signal received by said receiving antenna ;
the speech decoding apparatus as claimed in claim 13 that is adapted to
perform decoding processing on information obtained by said RF
demodulation apparatus;
a D/A conversion apparatus that is adapted to perform D/A conversion of
a digital speech signal decoded by said speech decoding apparatus ; and
a speech output apparatus that is adapted to convert an electrical signal
output from said D/A conversion apparatus to a speech signal.

The invention relates to an adaptive sound source vector generator (ASSVG)
(103) sets pitch cycles below and above the pitch cycle TO of integer accuracy
selected during the last sub-frame as a range for searching for a pitch frequency
of fraction accuracy and extracts an adaptive sound source vector P(T-frac)
having a pitch frequency T-frac of fraction accuracy in the range from an
adaptive codebook (ACB) (102). A last sub-frame integer pitch cycle storage
(LSFIPCS) (108) stores the integer component TO of the optimum pitch cyle
selected by a distortion comparator (DC) (107) and outputs the integer
component TO of the optimum pitch cycle to the adaptive sound source vector
generator (ASSVG) (103) when searching for the pitch cycle of the next sub-
frame. An optimum pitch frequency accuracy judge section (OPCAJS) (109)
judges whether the optimum pitch frequency is of integer accuracy or of fraction
accuracy. A comparison judge section (CJS) (110) limits the number of selections
of pitch information of fraction accuracy as the optimum pitch cycle.