Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Reexamination Certificate
1999-05-24
2001-12-04
Tsang, Fan (Department: 2645)
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
C704S201000, C704S220000
Reexamination Certificate
active
06327562
ABSTRACT:
FIELD OF THE INVENTION
The invention involves a procedure and a device for coding an audio-frequency signal, such as a speech signal, by means of “forward” and “backward” LPC analysis.
At present, the aim of coding techniques for audio-frequency signals, particularly speech signals, is to allow for the transmission of these signals in digital form, within the conditions of reduction of the transmission output, in order, particularly, to ensure a management adapted to the networks for transmitting these signals, taking into consideration the considerable growth in transactions between users.
BACKGROUND OF THE INVENTION
Of the coding techniques used, that designated by LPC analysis, Linear Predictive Coding in English, consists of carrying out a linear prediction of the audio-frequency signal to be encoded, the coding being carried out temporarily by means of a linear filtering prediction applied to the successive blocks of this signal.
Of the aforementioned techniques, that known as CELP coding, Code Excited Linear Prediction, is the most widespread and provides some of the best performance. Other techniques, such as the technique designated by MP-LPC, Multi Pulse Linear Predictive Coding, or the VSELP technique, Vector Sum Excited Linear Prediction in English, are relatively similar to CELP coding.
The aforementioned coding techniques are known as “analysis by synthesis”. They have enabled in particular, for audio-frequency signals belonging to the telephonic frequency bandwidth, the transmission output of these signals to be reduced from 64 kb/s (MIC coding) to 16 kb/s with the help of the CELP coding technique and even to 8 kb/s where these encoders use the most recent developments of this coding technique, without any perceptible reduction in the quality of the voice reconstituted after transmission and decoding.
A particularly important area of application for these coding techniques is, in particular, that of mobile telephony. Within this area of application, the necessary limitation of the frequency bandwidth granted to each mobile-telephony operator and the extremely rapid increase in the number of subscribers makes necessary the corresponding reduction of the coding output, while user demands in terms of speech quality continue to grow. Other areas of application of these coding techniques concern, for example, the storage of digital data which represent these signals on memory supports, high-quality telephony for video or audio conference applications, multimedia or digital transmissions via satellite.
The linear prediction filters used in the aforementioned techniques are obtained with the help of an analysis module called “LPC analysis” operating on successive digital signal blocks. These filters are capable, according to the order of analysis, that is, according to the number of filter coefficients, of modeling more or less reliably the contours of the spectrum of frequencies of the signal to be coded. In the case of a speech signal, these contours are called formants.
However, for good quality coding, required by most current applications, the filter thus defined is not sufficient for perfectly modeling the signal. It is therefore essential to code the residue of the linear prediction. One such operating mode relating to linear prediction residue is particularly used by the coding technique, LD-CELP, Low Delay CELP in English, previously mentioned in the description. In this case, the residual signal is modeled by a waveform taken from a stochastic codepage and multiplied by a gain value. The MP-LPC coding technique, for example, models this residue with the help of variable position pulses modified by respective gain values, whereas the VSELP coding technique carries out this modeling by means of a linear combination of pulse vectors taken from appropriate lists.
An explanatory recap of the operating method of LPC analysis and especially “backward” LPC analysis and “forward” LPC analysis will be given below.
The general envelope of the frequency spectrum is modeled by means of a short-term synthesis filter, constituting the LPC filter, the coefficients of which are modeled by means of a linear prediction of the speech signal to be coded. This LPC filter, an autoregressive filter, has a transfer function of the form, equation (1):
A
(
z
)
=
1
-
∑
i
=
1
p
a
i
z
-
i
where p designates the name of coefficients, ai of the filter and the order of the linear prediction applied, z designating the transformed variable z of the space of the frequencies.
One method of evaluating the coefficients a
i
consists of applying a criterion of minimization of the energy of the error prediction signal of the speech signal over the analysis length of this latter.
The analysis length for a digital speech signal formed of successive samples is, in practical terms, a number N of these samples, constituting a coding frame. The energy of the error prediction signal thus confirms equation (2):
Ep
=
∑
n
=
1
N
(
s
(
n
)
-
∑
i
=
1
p
a
i
·
s
(
n
-
1
)
)
2
where s(n) designates the sample of row n in the frame of N samples.
In a block-by-block coding process, the coding frame can be advantageously divided into several subframes or adjacent LPC blocks. The analysis length N then exceeds the length of each block in order to make it possible to take into account a certain number of past or, if applicable, future samples, by means of and at the cost of delaying the appropriate coding.
The analysis is called “forward” LPC when the LPC analysis process is carried out on the block of the current frame of the speech signal to be coded, with the coding taking place at encoder level “in real time”, that is, during the block of the current frame, with the only processing delay introduced by the calculation of the filter coefficients. This analysis involves transmitting the calculated values of the filter coefficients to the decoder.
“Backward” LPC analysis, used in the LD-CELP encoder at 16 kb/s is the object of the standard UIT-T G728. This analysis technique consists of carrying out the LPC analysis not on the block of the current frame of the speech signal to be coded, but on the synthesis signal. It is understood that this LPC analysis is actually performed on the synthesis signal of the block preceding the current block, as this signal is available simultaneously at encoder and decoder level. This simultaneous operation in the encoder and decoder thus makes it possible to avoid transmitting from the encoder to the decoder the value obtained in the encoder of the LPC filter coefficients. For this reason, “backward” LPC analysis makes it possible to free up transmission output and the output thus freed can be used, for example to enrich the excitation codepages in the case of CELP coding. “Backward” LPC analysis furthermore allows an increase in the order of analysis; the number of LPC filter coefficients may be as much as 50 in the case of an LD-CELP encoder, compared to 10 coefficients for most encoders using “forward” LPC analysis.
Thus, correct operation of “backward” LPC analysis requires the following conditions:
good quality synthesis signal, very close to the speech signal to be coded, which involves a sufficiently high coding output, higher than 13 kb/s, taking into account the quality of current CELP encoders;
reduced frame and block length due to the delay of one block between the analyzed signal and the signal to be coded. The length of the frame and block should therefore be low in comparison to the mean stationary time of the speech signal to be coded;
reliability of the transmission and conservation of the integrity of the data transmitted between the encoder and the decoder, by introducing few transmission errors. As soon as the synthesis signals differ significantly from the speech signal to be coded, the encoder and decoder cease to calculate the same filter and large divergences may occur, without being able to return to a noticeable similarity of the filters calculated in the encoder or decoder.
Due to the respective advant
France Telecom
Marshall Gerstein & Borun
Opsasnick Michael N.
Tsang Fan
LandOfFree
Method and device for coding an audio signal by... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and device for coding an audio signal by..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and device for coding an audio signal by... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2589421