Speech coding apparatus, linear prediction coefficient...

Details Speech coding apparatus, linear prediction coefficient... Speech coding apparatus, linear prediction coefficient...

1999-12-30

2001-03-20

Hudspeth, David R. (Department: 2741)

Data processing: speech signal processing, linguistics, language

Speech signal processing

For storage or transmission

C704S227000, C704S228000, C704S229000

Reexamination Certificate

active

06205421

ABSTRACT:

BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a speech coding apparatus in which pieces of speech information are coded to digital signals having a small information volume and the digital signals are transmitted and decoded to perform an efficient data transmission. Also, the present invention relates to a linear prediction coefficient analyzing apparatus in which a digital speech signal having an analyzing time-length is analyzed to obtain a linear prediction coefficient used in the speech coding apparatus. Also, the present invention relates to a noise reducing apparatus in which noise existing in speech information is reduced at a moderate degree before the speech information is coded in the speech coding apparatus.
2. Description of the Related Art
In a digital moving communication field such as a portable telephone, a compression coding method for speech signals transmitted at a low bit rate is required because subscribers in a digital moving communication have been increased, and research and development on the compression coding method have been carried out in various research facilities. In Japan, a coding method called a vector sum excited linear prediction (VSELP), proposed by the Motolora company, in which signals are transmitted at a bit rate of 11.2 kbits per second (kbps) is adopted as a standard coding method for a digital portable telephone. The digital portable telephone manufactured according to the VSELP coding method has been put on sale in Japan since the autumn of 1994. Also, another coding method called a pitch synchronous innovation code exited linear prediction (PSI-CELP), proposed by the NTT moving communication network Co., LTD., in which signals are transmitted at a bit rate of 5.6 kbps is adopted in Japan as a next standard coding method for a next portable telephone, and the development of the next portable telephone is going on now. These standard coding methods are obtained by improving a CELP which is disclosed by M. R. Schroeder in “High Quality Speech at Low Bit Rates” Proc. ICASSP, '85, pp.937-940. In this CELP coding method, speech information obtained from an input speech is separated into sound source information based on vibrational sounds of vocal cords and vocal tract information based on shapes of a vocal tract extending from the vocal cords to a mouth. The sound source information is coded according to a plurality of sound source samples stored in a code book while considering the vocal tract information and is compared with the input speech, and the vocal tract information is coded with a linear prediction coefficient. That is, an analysis by synthesis (A-b-S) method is adopted in the CELP coding method.
2.1. PREVIOUSLY PROPOSED ART
A fundamental algorithm of the CELP coding method is described.
FIG. 1
is a functional block diagram of a conventional speech coding apparatus according to the CELP coding method.
In
FIG. 1
, when a voice or speech is given to an input speech receiving unit
102
of a conventional speech coding apparatus
101
as pieces of speech data, an auto-correlation analysis and a linear prediction coefficient analysis for each of the speech data are performed in a linear prediction coefficient (LPC) analyzing unit
103
to obtain a linear prediction coefficient for each of the speech data. Thereafter, in the unit
103
, each of the linear prediction coefficients is coded to obtain an LPC code, and the LPC code is decoded to obtain a reproduced linear prediction coefficient.
Thereafter, all of first sound source samples stored in an adaptive code book
104
and all of second sound source samples stored in a probabilistic code book
105
are taken out to an adding unit
106
. In the adding unit
106
, an optimum gain for each of the first and second sound source samples is calculated, the sound source samples are power-adjusted according to the optimum gains, and a plurality of synthesis sound sources are obtained as a result of all combinations of the power-adjusted first sound source samples and the power-adjusted second sound source samples. That is, each of the synthesis sound sources is obtained by adding one of the power-adjusted first sound source samples and one of the power-adjusted second sound source samples.
Thereafter, in an LPC synthesizing unit
107
, the synthesis sound sources are filtered with the reproduced linear prediction coefficient obtained in the LPC analyzing unit
103
to obtain a plurality of synthesis speeches. Thereafter, in a comparing unit
108
, a distance between each of the speech data received in the input speech receiving unit
102
and each of the synthesis speeches is calculated, a particular synthesis speech corresponding to a particular distance which is the minimum value among the distances is selected from the synthesis speeches, and a particular first sound source sample and a particular second sound source sample corresponding to the particular synthesis speech are obtained.
Thereafter, in a parameter coding unit
109
, the optimum gains calculated in the adding unit
106
are coded to obtain a plurality of gain codes. The LPC code obtained in the LPC analyzing unit
103
, index codes indicating the particular sound source samples obtained in the comparing unit
108
and the gain codes are transmitted to a transmission line
110
in a group. Also, a synthesis sound source is generated from a gain code corresponding to the particular first sound source sample and the particular first sound source sample in the unit
109
. The synthesis sound source is stored in the adaptive code book
104
as a first sound source sample, and the particular first sound source sample is abandoned.
In addition, in the LPC synthesizing unit
107
, acoustic feeling for each of the speech data is weighted with the linear prediction coefficient, a frequency emphasizing filter coefficient and a long-term prediction coefficient obtained by performing a long-term prediction analysis for each of the speech data. Also, the sound source samples are found out from sub-frames obtained by dividing each of analyzing blocks in the adaptive code book
104
and the probabilistic code book
105
.
Also, the linear prediction coefficient analysis performed in the LPC analyzing unit
103
is utilized in various coding methods. A conventional linear prediction coefficient analysis is described with reference to FIG.
2
.
FIG. 2
is a block diagram of a conventional linear prediction coefficient analyzing apparatus.
As shown in
FIG. 2
, when a speech is input to an input speech receiving unit
112
of a conventional linear prediction coefficient analyzing apparatus
111
, the speech is converted into a plurality of speech signals Xi respectively having a prescribed analyzing period, and each of the speech signals Xi output time-sequentially is multiplied by a window coefficient Wi in a window putting unit
113
. For example, a coefficient in a Hamming window, a Hanning window, a Blackman-Harris window or the like is used as the window coefficient Wi. A window putting processing in the unit
113
is formulated as follows.
Yi=Wi*Xi
Here, i denotes the numbers of the speech signals (i=1 to L), L denotes the number of speech signals, and Yi denotes a plurality of window-processed speech signals.
Thereafter, an auto-correlation analysis is performed for the window-processed speech signals Yi in an auto-correlation analyzing unit
114
as follows.
Vj
=
∑
i
=
j
+
1
L
⁢
Yi
*
Yi
-
j
Here, Vj denotes a plurality of auto-correlation functions, and j denotes the numbers of the auto-correlation functions.
Thereafter, a linear prediction analysis based on an auto-correlation method is performed in a linear prediction coefficient analyzing unit
115
to obtain a linear prediction coefficient for each of the speech signals. The linear prediction analysis is disclosed in various speech information processing documents such as “The Autocorrelation Method” in a literature written by L. R. Labiner and R. W. Schafer “Digital Processing of Speech Signals” pp.401-

Affiliated with

Also associated with

Rating

Speech coding apparatus, linear prediction coefficient... does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Speech coding apparatus, linear prediction coefficient..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Speech coding apparatus, linear prediction coefficient... will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2497737