Research On Wideband And Variable Rate Speech Coder Based On SILK

To provide users with more comfortable listening experience,wider bandwidth and higher sampling frequency are required to improve the quality of speech coding.Since SILK can provide both wideband variable rate speech coding and better call quality in a low-bandwidth situation,the application prospects of SILK are widely concerned.It has important research significance and application value to design a high quality speech coder with a SILK based wideband variable rate and apply it to real-time speech communication scenarios.In this thesis,the key algorithm and performance enhancement of SILK are investigated.At present,the SILK is mostly used in Voice over Internet Protocol(VoIP)applications where the Internet is used as a transport bearer.Since the Internet only provides a best-effort service,the packet loss often occurs due to network delay,congestion,and error propagation,which seriously affect the quality of voice at the receiving end.To effectively solve the packet loss problem in VoIP,this thesis conducts an research on packet loss processing technology based on SILK speech encoder with wideband variable rate,and an Improved Low Bit Rate Redundancy coding(ILBRR)combined with Interpolation algorithm is proposed.The quality of synthesized speech of the standard SILK and the SILK using the I-ILBRR algorithm are tested under different packet loss rates.Experimental results show that the SILK encoder using the I-ILBRR algorithm is more fault tolerant.Since adopting the I-ILBRR algorithm will increase the average encoding rate of SILK speech coder,this thesis simulates the input signals and proposes a Predictive Noise Shaping Quantizer(PNSQ)algorithm to improve the SILK coding efficiency.Firstly,add the predetermined noise to the input speech signals so that the encoder side can generate a simulated signal that matches the spectral characteristics.Then,combine the long-term prediction and the short-term prediction of the simulated signal to increase the prediction gain of the prediction filter and reduce the entropy of quantization index,which can reduce the required bit rate to transmit the encoded speech signal.In addition,the SILK encoder that uses the PNSQ algorithm can improve coding efficiency by neglecting the requirement of additional sideband information and the change of bitstream format.The test results show that the proposed algorithm reduces the average encoding bit rate by more than 1.5223 kbps while the quality of synthesized voice is basically unchanged.