[Voice coding] based on matlab ADPCM coding and decoding [including Matlab source code 553 period]

1. Introduction

1 Adpcm coding principle
Insert picture description here
Coding steps:

Calculate the difference diff between the input pcm data and the predicted pcm data (the last pcm data for the first time); calculate the delta through the differential quantizer (by index (the first encoding index is 0) to find the step, through diff and step Find the delta). The delta is the encoded data;
the vpdiff is calculated by the inverse quantizer (the vpdiff is calculated by the calculated delta and step);
the new predicted valpred is calculated, that is, the last predicted valpred+vpdiff; the
predictor (normalized ), find the predicted pcm value of the current input pcm input, for the next calculation;
quantization step adjustment (by delta look-up table and index, calculate the new index value). For the next calculation;
2 adpcm decoding principle
Insert picture description here
Decoding steps (in fact, the decoding principle is the third to sixth steps of encoding):

Find vpdiff through the inverse quantizer (from the stored delta and index, find the step, calculate the vpdiff);
find the new predicted valpred, that is, the last predicted valpred+vpdiff;
through the predictor (normalized), find The predicted pcm value of the current input pcm input is used for the next calculation. The predicted pcm value is the decoded data; the
quantization step is adjusted (the new index value is calculated by looking up the table and index through delta). For the next calculation;
note description
Through the principle of encoding and decoding, we can see that in fact, the first encoding has already been decoded, that is, the predicted pcm.
Because the output data after encoding and decoding has been quantized. According to the calculation formula delta = diff*4/step; vpdiff = (delta+0.5)*step/4; Considering that they are all integer operations, it can be derived: the predicted pcm data generated by encoding and then decoding the pcm data, if the pcm data is predicted The data obtained by re-encoding is the same as the data obtained by the first encoding. Therefore, after the pcm data is lost after one encoding, no matter how many times it is decoded and re-encoded, the data is the same, and the sound quality will not be lost again. That is, after the first encoding, no matter how many times the data is encoded and decoded, it is a lossless output.

3 ADPCM data storage format
This part is the description of adpcm data storage, which belongs to the detail part. Many codes are noisy when decoded because the details of this part are incorrect, so you need to read it carefully.

3.1 Introduction to adpcm data block
Adpcm data is stored in one block and one block, and the block is composed of both block header (block header) and data. The block header is a structure, and its definition in mono is as follows:

Typedef struct
{
    
    
short  sample0;    //block中第一个采样值（未压缩）
BYTE  index;     //上一个block最后一个index，第一个block的index=0;
BYTE  reserved;   //尚未使用
}MonoBlockHeader;

For dual channels, its blockheader should contain two MonoBlockHeaders, which are defined as follows:

typedaf struct
{
    
    
MonoBlockHeader leftbher;
MonoBlockHeader rightbher;
}StereoBlockHeader;

When decompressing, the left and right channels are processed separately, so there must be two MonoBlockHeaders;
with the blockheader information, you can easily extract the compressed data in this block without knowing the previous data of the block. Therefore, adpcm decoding is only related to this block and has nothing to do with other blocks. It can only decode any block data individually.
The size of the block is fixed and can be customized. The number of samples nsamples contained in each block is calculated as follows:

//
#define BLKSIZE 1024
block = BLKSIZE * channels;
//block = BLKSIZE;//ffmpeg
nsamples = (block  - 4 * channels) * 8 / (4 * channels) + 1;

For example, the audition software uses the above, a single-channel block is 1024 bytes, 2041 samples, and a dual-channel block is 2048, which also contains 2041 samples.
And ffmpeg uses block=1024bytes, that is, both single and double channels are 1024bytes, and the number of samples for single and double channels can be calculated by the formula to be 2041 and 1017 respectively;

3.2 Single-channel pcm format:
Insert picture description here
3.3 Dual-channel pcm format:

3.4 Implementation of encoding and decoding codes
Special attention should be paid to the processing of two channels and the processing method when the data is not enough for 1 block. The
code includes encoding and decoding test cases, which implement encoding and then decoding. Welcome to exchange and learn the
complete code download address (this article only details the adpcm encoding and decoding, if you want the wav file encoding and decoding to be correct, you need to download the complete code. The
complete code is 0x0011 /* Intel's DVI ADPCM */ encoding and decoding code implementation. Including single and double channels The processing and the final data are not the processing of the whole block):

Second, the source code

clear all;
clc
close all;
[x,fs,numbits]= wavread('C6_3_y.wav'); 

sign_bit=2;                                     %两位ADPCM算法
ss=adpcm_encoder(x,sign_bit);
yy=adpcm_decoder(ss,sign_bit)';

nq=sum((x-yy).*(x-yy))/length(x);
sq=mean(yy.^2);
snr=(sq/nq);
t=(1:length(x))/fs;
subplot(211)
plot(t,x/max(abs(x)))
axis tight
title('(a)编码前语音')
xlabel('时间/s')
ylabel('幅度')
subplot(212)
plot(t,yy/max(abs(yy)))
axis tight
% APDCM解码函数
function y=adpcm_decoder(code,sign_bit)
len=length(code);
y = zeros(1,len);
ss2 = zeros(1,len); 
ss2(1) = 1; 

currentIndex =1; 
index = [-1 4]; 
startval = 1; 
endval = 127;
base = exp( log(2)/8 ); 
% 近似步长 
const = startval/base; 
numSteps = round( log(endval/const) / log(base) ); 
n = 1:numSteps; 
base = exp( log(endval/startval) / (numSteps-1) ); 
const = startval/base; 
table2 = round( const*base.^n ); 
for n = 2:len 
% 计算量化距离 
    neg = code(n) >= sign_bit; 
    if (neg) 
        temp = code(n) - sign_bit; 
    else 
        temp = code(n); 
    end 
    temp2 = (temp+.5)*ss2(n-1); 
    if (neg) 
        temp2 = -temp2; 
    end

Three, running results

Insert picture description here

Four, remarks