| With the development of voice interaction technology and intelligent hardware,intelligent voice terminal devices are increasingly entering people's lives.In the vast majority of intelligent speech equipment,speech processing is the most important,and its performance will directly affect the user's auditory experience or recognition performance.The echo cancellation algorithm is one of the most important technical difficulties in the whole speech processing.It needs to eliminate the echo signal coupled by the speaker from the signal picked up by the microphone,and at the same time,it also needs to retain the speech of the near-end speaker as much as possible to improve the double talk performance.Based on the previous work and traditional algorithms,the embedded DSP hardware is used as the algorithm verification platform,and the methods of theoretical modeling,software simulation and software/hardware co-tuning are used to deeply study the echo cancellation scheme based on adaptive filtering and the echo suppression scheme based on spectrum correction technology,respectively.Meanwhile,for wide-band intelligent voice terminal products,considering the computational complexity and echo cancellation performance,a sub-band echo cancellation scheme based on AEC and AES cooperation is proposed,and some research results have been obtained.The main research work is as follows:1.The basic working principle of echo canceller and the mechanism of acoustic echo are studied,and the principles and mathematical implementation processes of two commonly used algorithms,namely echo cancellation algorithm based on adaptive filtering and echo suppression algorithm based on spectrum modification technology,are studied and analyzed respectively,and the quality evaluation methods of commonly used echo cancellers are summarized;2.An improved adaptive echo cancellation algorithm based on frequency domain optimal variable step size is proposed.Firstly,the optimal step theory in adaptive filter is studied,and then FDAF algorithm based on state space model is studied in detail.Secondly,a decision factor for detecting abrupt changes in acoustic paths is emphatically studied,and the decision factor is mapped into model parameters by non-linear function.Thirdly,steady-state analysis is used to explain the defects of the original algorithm and the effectiveness of the proposed algorithm.Finally,the performance between the proposed algorithm and the original algorithm is verified by experiments.3.A low complexity echo suppression algorithm based on Wiener filter is proposed.Firstly,the parametric Wiener filter based on prior SER ratio estimation is studied,and a parameter adaptive strategy based on posterior SER and double-talk decision factor is proposed.The simulation results show that the proposed algorithm can effectively improve the echo suppression degree and guarantee the speech quality at the same time.4.A wide-band echo cancellation algorithm based on AEC and AES cooperation is proposed.Firstly,the two-channel orthogonal filter banks are studied.Secondly,the low frequency band echo cancellation algorithm based on block State-Space FDAF and the full frequency band post echo suppression algorithm based on Wiener filter are studied respectively.Simulation results show that the proposed algorithm can complete echo cancellation of wide-band audio system with low computational complexity.5.Design and implement a speech communication system based on embedded DSP platform.On the basis of completing algorithm simulation and obtaining good results,the audio algorithm engine is designed for wide-band speech communication scenarios with the self-developed TI-C6748 DSP hardware platform.Finally,a set of software and hardware system for real-time speech communication is realized.Experimental results show that the system can meet the requirements of conventional real-time speech communication and can be well applied to intelligent voice terminal products based on TI-C67 x DSP. |
PDF Full Text Request