Sunteți pe pagina 1din 13

RELP

(RESIDUAL EXCITED LINEAR PREDICTION)

Introduction
LPC ---> Low bit-rate speech coding ---> Toll quality speech ---> Compression of more speech channels within a given BW Signal Compression : Transmission of speech with the highest possible quality using the least channel capacity i.e. higher spectral transmission efficiency. Speech coders : Waveform coders ; Vocoders (RELP, CELP) Vocoders consists of 3 basic steps 1. Analyzing the voice signal and determination of spectral parameters at the transmitter 2. Transmission of the spectral parameters 3. Synthesis of the speech at the receiver end using these parameters

The RELP vocoder, which Un et al. proposed, encodes speech between 6 and 9.6 kbps, depending on the quality of the synthesized speech desired. The total encoding rate of the RELP vocoder is larger than most of the other LPC-based vocoder systems. Linear prediction error (residual) signals are used for the excitation. This improves the quality of the synthesized speech and makes it more natural than the basic LPC vocoders, because there are no misclassification of voiced/unvoiced sounds or miscalculation of pitches. The RELP vocoder needs to encode sequences of residual signals per segment, which is a large volume of data, and hence, its compression rate is moderate while several bits are needed to encode the voiced/unvoiced decision, pitch, and gain for the other LPC systems.

RELP system conceptualization


Number of pre-determined spectral coefficients : L

+
s(n)

h -

e(n)

h : spectral coefficients Synthesized speech signal is obtained by s(n) = s(n) x h Residual sequence : e(n) = s(n) s(n)

Residual sequence is passed through a LPF(low-pass filter) inorder to extract a baseband signal(BBS) from it. Decimation of BBS : The decimation factor depends on how much compression is desired and on how much aliasing is perceptually acceptable in the baseband of the output speech. BB residual sequence , LPC coefficients and first L original speech values are coded to bits and transmitted.
Transmitter part of RELP system

Receiver side Received signal, r, interpolated back to its original sampling rate. Received signal is full-band reconstructed through non-linear distortion. BBS intact ; only high frequencies to be reconstructed. Interpolated signal is full-wave rectified. Resulting signal is high-pass filtered using the same cut-off frequency as the low-pass filter. Now the baseband interpolated signal is added with the highfrequency signal to produce the residual sequence.

Reconstructed speech signal is calculated using : =yxh+e

Receiver part of RELP system

RELP (proposed by Un et al.)


Consists of five functional blocks:

LPC analyzer, Residual encoder Residual decoder Spectral flattener LPC synthesizer.

Block diagram of RELP vocoder

Speech signal is Low-pass filtered : Cut-off frequency = 3.2 kHz Analog-to-digital conversion : Sampling Rate = 6.8 kHz Speech samples : Hamming window (Length = 196 sample points)
196 sample points

136 sample points : specified as a 20-ms analysis block

30 samples each of previous block and next block : added to analysis block for the overlaps

These overlaps ensure smooth transitions to avoid abrupt changes between analysis blocks. Adaptive Delta Modulation (ADM) encoder : used for the residual coding. The residual signals are low-pass filtered at 800 Hz to reduce the transmission rate before coding ie. Low Bit-Rate Coding. At the residual decoding process, the reproduced residual signals are generated from the transmitted sign bits through the ADM decoder. The output signals of the ADM decoder are low-pass filtered for the purpose of smoothing them. Spectral flattener : Recover the high- frequency harmonics of the residual signals. The synthesized speech then is generated by the LPC synthesizer, with the recovered residual as the excitation signal.

Conclusions derived
Total transmission rate : 9600 bps is comprised of Residual signal transmission rate : 6800 bps Coefficients : 2250 bps Gain : 200 bps Normalized energy : 200 bps Frame synchronization : 150 bps Sampling rate : 6.8 kHz Order of LPC filter : 10

At this rate, the synthesized speech is intelligible and the speaker can be easily identified.

THE END

S-ar putea să vă placă și