Adaptive postfilter

Details Adaptive postfilter Adaptive postfilter

1999-01-20

2003-06-24

Abebe, Daniel (Department: 2741)

Data processing: speech signal processing, linguistics, language

Audio signal bandwidth compression or expansion

C704S200000, C704S219000

Reexamination Certificate

active

06584441

ABSTRACT:

RANGE OF THE INVENTION
The invention relates to the coding of speech at variable bit rates, whereby the bit rates can vary from frame to frame, and more specifically to the methods and filters used for improving the quality of decoded speech.
BACKGROUND OF THE INVENTION
The coding of speech at a variable bit rate can be used to maximize the capacity of a data transfer connection at a certain level of speech quality, or to minimize the average bit rate of a speech connection. This is possible because speech is not homogeneous, and if speech is divided into short sections, different sections can be presented using a different number of bits in each section without a perceivable difference in quality. Codecs using a fixed bit rate must operate at a kind of compromise rate, which is not too high in order to save data transfer capacity, but high enough to present different parts of speech with sufficient quality. This compromise rate is needlessly high for the sounds that could be presented with a smaller number of bits. The variable-rate method of speech coding can be used to advantage in many applications. Packet-switched networks, such as internet, can use variable-rate communications directly by sending different sized packages. The Code Division Multiple Access (CDMA) systems can also directly utilize variable-rate coding. In the CDMA systems, the average fall of the transmission rate reduces the mutual disturbances caused by different transmissions and makes it possible to increase the number of users. In the so-called third generation mobile station systems, variable-rate data transfer is likely to be used in some form. In addition to data transfer, variable-rate coding is also useful in connection with voice recording and voice message systems, such as telephone answering machines, where the saving due to variable-rate coding is seen as saved recording capacity.
The bit rate of a variable-rate codec can be controlled in many ways. One way is based on monitoring the capacity of the data transfer network, whereby the momentary bit rate is determined according to the available capacity. In a system like this, the bit rate can also be set an upper and lower limit on the basis of the capacity in use. The limits of the capacity are seen as reduced speech quality particularly during times of congestion, when the system forces the bit rate down.
Variable-rate coding can also be used to implement an error-tolerant coding method for mobile stations. In a method like this, the bit rate of speech coding is adapted on the basis of the quality of the transmission channel. When the quality of the transmission channel is good, the bit rate is kept relatively high and in addition to the coded speech only a little error correction information is transferred. In good transmission conditions, this method is sufficient to remove transmission errors. When the quality of the transmission channel becomes worse, the bit rate is lowered, whereby stronger channel coding can be used in an ordinary fixed-rate transmission channel. Then the reduction of speech quality is minimized by means of this stronger channel coding, which can correct larger errors. However, speech quality is reduced somewhat when the quality of the transmission connection is weakened, because the bit rate is lowered.
A typical CELP coder (Code Excited Linear Prediction) comprises many filters modelling speech formation, for which a suitable excitation signal is selected from the excitation vectors contained by the codebook. A CELP coder includes typically both short-term and long-term filters, in which a synthesized version of the original speech signal is formed by filtering excitations selected from the codebook. An excitation vector producing the optimum excitation signal is sought from the excitation vectors of the codebook. During the search, each excitation vector is applied to the synthesizer, which includes both short-term and long-term filters. The synthesized speech signal is compared to the original speech signal, taking account of the response of the human hearing capacity, whereby a characteristic comparable to the observed speech quality is obtained. An optimum excitation vector is obtained for each part of the speech signal being processed by selecting from the codebook the excitation vector which produces the smallest weighted error signal for the part of the speech signal in question. CELP coders like this are described in more detail in the patent specification U.S. Pat. No. 5,327,519, for instance.
FIG. 1
shows an example of a block diagram of a prior art fixed-rate CELP coder. The coder comprises two analysis blocks, namely the short-term analysis block
10
and the long-term analysis block
11
. These analyse the speech signal s(n) to be coded, the short-term analysis block mostly the formants of the spectrum of the speech signal and the long-term analysis block mostly the periodicity (pitch) of the speech signal. The blocks form multiplier sets a(i) and b(i), which determine the filtering properties of the short-term and long-term filter blocks. The multiplier set a(i) formed by the short-term analysis block corresponds to the formants of the spectrum of the speech signal to be coded, and the multiplier set b(i) formed by the long-term analysis block corresponds to the periodicity (pitch) of the speech signal to be coded. The multiplier sets a(i) and b(i) are sent to the receiver through the data transfer channel
5
. The multiplier sets are calculated separately for each frame of the speech signal to be coded, the temporal length of the frames being typically 20 ms.
The long and short-term filter blocks
13
,
12
filter excitations selected from the codebook according to the multiplier sets a(i) and b(i). The long-term filter thus models the periodicity (pitch) of the voice, or the vibration of the vocal cords, and the short-term filter models the formants of the spectrum, or the human voice formation channels. The filtering result ss(n) is reduced from the speech signal s(n) to be coded in the summing device
18
. The residual signal e(n) is taken to the weighting filter
14
. The properties of the weighting filter are chosen according to the human hearing capacity. The weighting filter attenuates the frequencies which are perceptually less important, and emphasizes those frequencies which have a substantial effect on the perceived speech quality. The code vector search control block
15
searches on the basis of the output signal of the weighting filter a corresponding excitation vector index u. The excitation codebook
16
forms the desired excitation on the basis of the code vector corresponding to the index, and the excitation is fed to the multiplication device
17
. The multiplication device forms the product of the excitation and the weighting factor g of the excitation given by the code vector search control block, which product is fed to the filter blocks
12
,
13
. The code vector search control block searches iteratively for an optimum excitation code vector. When the residue signal e(n) is at the minimum or sufficiently small, the desired code vector is considered to be found, whereby the index u of the excitation code vector and the weighting factor g are sent to the receiver.
FIG. 2
shows an example of a block diagram of a prior art CELP decoder. The decoder receives the coding parameter sets a(i) and b(i), the weighting factor g and the excitation code vector index u from the data transfer channel
5
. An excitation code vector corresponding to the index u is selected from the excitation codebook, and a corresponding excitation c(n) is multiplied in the multiplication device
21
with the weighting factor g. The resulting signal is fed to the long-term synthesizing filter
22
and further to the short-term synthesizing filter
23
. The coding parameter sets a(i) and b(i) control the filters
22
,
23
in the same way as in the coder of FIG.
1
. The output signal of the short-term filter is filtered further in a postfilter
24
for forming a reconstructed speech signal s′(n).
In a

Affiliated with

Also associated with

Rating

Adaptive postfilter does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Adaptive postfilter, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Adaptive postfilter will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3153566