A Number Of Voice, Audio Signal Processing Algorithms Research And Realization

Speech is one of the most basic and important means of communication. As an important component of speech communication, narrow-band speech coding has been researched for several decades, and a series of standards on speech coding have been established, i.e. G.728, G.723, etc. Recently, the wide-band audio coding is also developing rapidly, which aims to satisfy various requirements such as personal communications, internet multimedia and digital broadcast etc. Due to the development of the multi-channel audio coding and the virtual sound technology, they have got broad application along with the steps of Digital Times. The main content of this thesis includes the following three aspects: 1. The study on LD-CELP (Low-Delay Code Excited Linear Prediction) algorithm of G.728 codec, and the porting techniques on the fixed-point DSP chip; 2. The research on the multi-channel audio codec and the fixed-point implementation methods; 3. The system design of the virtual reproduction of AC-3 5.1 channel surround sound by loudspeaker.Firstly, based on the study of LD-CELP algorithm of the G.728 codec, which is a float-point algorithm, this thesis focuses on the implementation of the fixed-point G.728 algorithm and the porting techniques on the DSP chip, ADSP-BF533. Furthermore, this thesis summarizes some fixed-point techniques, which are helpful for the porting of other float-point algorithms to the fixed-point DSP chip. The subjective hearing tests have shown that the fixed version on BF533 has almost the same effect as the original G.728.Secondly, the representative techniques of multi-channel audio codec, including MPEG AAC, Dolby AC-3 and DTS etc., are studied in this thesis. And the AC-3 and DTS decoders are fixed-pointed according to the experiences from the previous work. Although there are little differences on the waveform, the results of the subjective hearing tests are promising.Finally, we design the system of virtual reproduction of AC-3 5.1 channel surround sound by combing the virtual reproduction of 5.1 channel techniques and Dolby AC-3 decoder, in which the subband-domain filtering method in MDCT domain is adopted in the virtual processing of 5.1 channel signals from AC-3 decoder. Compared with the general time-domain filtering methods, this method can greatly reduce the computation load with the equivalent 3D-sound hearing effect, which is important for the real-time realization. Furthermore, the method can also be applied for other multi-channel audio codec systems, such as MPEG AAC, DTS and so on.