CONTENT:
* AMR Introduction
* AMR Voice quality Assessment
* AMR File Structure analysis
* AMR Frame structure Analysis
* AMR frame reading algorithm
* AMR decoding principle and process
* AMR Mode selection Adaptive mechanism
I. Introduction of AMR
Based on the new network and new requirements, whether in terms of saving transmission frequency band resources or maintaining the efficiency of line communication, it is of great significance to study the system with various variable rate speech coding techniques. At present, in order to adapt to this need to put forward the concept of AMR (Adaptive multi-rate), that is, adaptive multi-rate voice encoder, mainly for mobile device audio, compression compared to larger, but compared with other compression format quality is poor, because more for voice calls. AMR is divided into two kinds, one is AMR-NB (Amr-narrowbind), the Voice bandwidth range: 300-3700hz,8khz sampling frequency, the other is AMR-WB (AMR wideband), the voice bandwidth range 50-7000hz, 16KHz sampling frequency. But considering the short-time correlation of speech, each frame length is 20ms. The two encoders have different rates depending on the bandwidth requirements, but there are similarities.
(1) AMR-NB
AMR has a sampling frequency of 8KHz, one frame per 20ms encoding, each frame contains 160 voice samples.
AMR is based on the generation of digital excitation linear prediction (ACELP) coding mode, the coding end of the extraction of ACELP model parameters (linear predictor, adaptive codebook and Fixed code this index and gain), the decoder side received the data and then the new synthesis of speech based on these parameters.
Implementation of AMR-NB in TD-SCDMA. This encoder uses the generation of digital-based linear prediction (ACELP) hybrid coding, that is, the digital voice signal includes both a number of speech features and part of the waveform encoding information, and then use these features information to re-synthesize the voice signal process. Control the number of extracts of these parameters, according to the rate requirements of the information to choose the following 8 rates, mixed composition as shown in table one of the adaptive speech Encoder. The pattern amr_12.20 extracts 244 bits of parameter information, while the pattern amr_4.70 extracts only 95 bits of information. Depending on the amount of information contained in these bits, it can be divided into 3 classes of bit class 0,1 and 2. In the channel encoding Class 0 and 1 will use cyclic redundancy check code for error detection, for Class 2 is based on the previous frame to recover.
Table One: AMR encoder encoding Rate
Encoding model |
Encoder bit rate |
Coded model |
Encoder bit rate |
amr_12.2 |
12,20kbit/s (GSM_EFR) |
amr_5.90 |
5,90 kbit/s |
AMR amr & nbsp |
10,20 kbit/s |
amr_5.15 |
5,15 kbit/s |
amr_7.95 |
7,95 kbit/s |
amr_4.75 |
&nb sp;4,75 kbit/s |
amr_7.40 |
7,40kbit/s (IS-641) |
Amr_sid |
1,80 kbit/s (no voice message transmission) |
amr_6.70 |
6,70kbit/s (PDC-EFR) |
|
|