Research Of AMR-WB Algorithm And DSP Implementation

To meet the requirements for high-quality voice service of modern communication,3 GPP proposed Adaptive Multi-Rate Wideband (AMR-WB) codec. Later, AMR-WB was selected as wideband speech coding standard around 16kbit/s named G.722.2 by ITU-T. The audio bandwidth of narrow-band speech is limited at 200Hz-3400Hz, and the sampling frequency is 8 kHz, while the audio bandwidth of the wideband speech extends to 50Hz-7kHz, and the sampling frequency is 16 kHz. Because of the bandwidth extention and the increase sampling rate, the codec performance in the understandability, naturalness and the transmission of speech has been greatly improved. AMR-WB gives superior voice quality, low average coding rate, and good adaptation. It is the first speech coding system can be used for both wired and wireless service in the communications history. This algorithm can be used in the fields like teleconference, multimedia communication and high-definition television and so on, so the DSP implementation of the algorithm has a broad application prospect.Firstly, AMR-WB algorithm is analyzed in detail on the linear prediction quantification, adaptive codebook searching, fixed codebook searching, the high frequency band processing and several other modules of the encoder and the decoding principles, including parameters decoding and speech synthesis, high-pass filter, the sampling and interpolation, and treatment of high frequency band of the decoder; then the algorithm is simulated in the environment of VC++6.0. Voice experiments are done using the standard TIMIT English speech database files, and the synthesized speech is compared in waveform, subjectively heard and objectively tested by w-PESQ.Secondly, the implementation of the algorithm on DSP TMS320VC5509A is mainly researched, and the work contains details as follows:(1) The hardware resources on the DSP TMS320VC5509A are well studied and researched for the initialization programming, control register configuration and the interface programming of the modules, such as voice interface chip, I2C bus, multi-channel buffered serial port (McBSP) and direct memory access (DMA) controller and external memory interface (EMIF), after this, the hardware platform of the AMR-WB algorithm has been constructed;(2) The software design of the system is completed by the means of the system initialization procedures, speech coding and decoding process and the design of audio interface module for receiving and sending speech frames, which are controlled by the two interrupt program. The purpose of the hardware and software design is to implement the algorithm;(3) The AMR-WB algorithms is transplanted to DSP, specifically including rewriting the Gel file, design the.map file, allocation of memory in C5509A, redefine the data type and resolving the compatibility problems between C5509A and PC;(4) The most significant work is the algorithm optimization, which can be divided into three parts:the optimization based on the optimizer, the optimization of basic operators and the optimization of key functions. Among these, the optimization based on the optimizer is the primary one refer to this algorithm by selecting proper optimizing options; the optimization of basic operators is mainly for, such as addition, subtraction, multiplication, division, shifting, rounding, saturation overflow, multiplication addition and reduction and so on; the main objects of the optimization of key functions are functions that consume the instruction cycles more and are called frequently in the encoding and decoding process, which are got through the analysis by the Profile tool in CCS;(5) The results of the codec has been tested and evaluated. After the standard testing input audio file supplied by 3GPP encoded and decoded, the output audio files can be obtained, which is tested with the standard testing output audio file, and subjective hearing test is done too.