A mixed and code excitation LPC vocoder at 1.76 kb/s

This paper describes a speech-coding algorithm that we develop recently. We call it M&CELP VOCODER. It is based on the traditional linear prediction vocoder. The excitation is constructed by not usually one of either periodic pulse or white noise, but the mixture of periodic feature-waveform (from predict error) and periodic pulse and white noise passed high pass filter in voiced speech section and white noise in the unvoiced speech. At parameter quantization, The quantizer based on Tree-search technique is applied, and it is improved. It enhanced the performance of reconstructed speech, while maintaining the lower bit rate. In addition, the scheme of pitch evaluation is improved, and this vocoder is very efficient to eliminate thumps and buzz, and it enhanced the naturalness and intelligibility. The algorithm is very efficient at speed, it can be realized at the PC computer with high configuration. Informal listening confirm that it is as good as the new U.S. federal standard: MELP at 2.4kb/s.

[1]  Juan Carlos De Martin,et al.  A 1.7 kb/s MELP coder with improved analysis and quantization , 1998, Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181).

[2]  Peter Kabal,et al.  The computation of line spectral frequencies using Chebyshev polynomials , 1986, IEEE Trans. Acoust. Speech Signal Process..

[3]  Joseph P. Campbell,et al.  Voiced/Unvoiced classification of speech with applications to the U.S. government LPC-10E algorithm , 1986, ICASSP '86. IEEE International Conference on Acoustics, Speech, and Signal Processing.

[4]  Yusuke Hiwasaki,et al.  A new 2-kbit/s speech coder based on normalized pitch waveform , 1997, 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[5]  Thomas P. Barnwell,et al.  A new mixed excitation LPC vocoder , 1991, [Proceedings] ICASSP 91: 1991 International Conference on Acoustics, Speech, and Signal Processing.