A low-bit-rate speech codec using multiband excitation and LPC modeling of harmonic magnitudes for private mobile radio
Teruo Fumoto, Seishi Sasaki, Koji Kondo · Electronics and Communications in Japan (Part II Electronics) · 2004
This paper discusses a low-bit-rate speech codec that can be operated in the 3.125-kHz band, which is half the minimum frequency band occupied at present, for use in next-generation private mobile radio. In the speech codec under consideration, the frame length is 20 ms (10 ms × 2 subframes), and the pitch, the harmonic magnitudes, and the voiced/unvoiced (V/UV) information are analyzed and extracted from each subframe. Then, the logarithmic pitch is uniform/prediction quantized. A linear model is constructed by correcting the DC component of the harmonic magnitude, and the harmonic magnitude is quantized separately for the prediction coefficient and the gain. The V/UV information is quantized by means of a representative pattern number. Thus, the speech is encoded with a coding rate of 1.6 kbit/s. Error correction is applied at 1.6 kbit/s. The coding scheme is divided into three classes according to the error sensitivity of the speech coding bits, and is encoded at different coding rate by combining CRC (cyclic redundancy check) and RCPC (rate compatible punctured convolutional code). Finally, clean speech is used and speech quality tests employing the MOS (mean opinion score) and the articulation test are performed in an error generation environment. It is shown that although the MOS under error-free conditions is a little worse than FS1016 (CELP, 4.8 kbit/s), the quality degradation is slow when errors are generated. Single sound articulation above 80% is obtained for an error rate of 7%, which is considered sufficient for speech communications. © 2004 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 88(1): 40–50, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjb.20140