5-kHz-bandwidth speech coder at 4-8 kbit/s

Abstract

We propose a 5-kHz-bandwidth CELP speech coder for various multimedia applications. For portability, the coder has three bit rate modes: MODE8 (7.8 kbit/s), MODE6 (5.75 kbit/s) and MODE4 (3.95 kbit/s). The bit rate mode can be switched frame by frame. In order to achieve both low bit rate and naturalness, 5-kHz-bandwidth speech signals are used instead of 3.4 kHz or 7-kHz-bandwidth signals. The speech signals under consideration are band-limited to 5 kHz and are sampled at 11.025 kHz. Subjective listening tests indicated that the 5-kHz-bandwidth is effective for low-bit-rate speech coders. The mean opinion score and comparative mean opinion score showed that the quality of this coder in MODE8 (5 kHz, 7.8 kbit/s) is better than that of the G.729 (3.4 kHz, 8 kbit/s), G.722 (7 kHz, 48 kbit/s), and equivalent to that of G.729 Annex E (3.4 kHz, 11.8 kbit/s). In addition, at MODE6 (5 kHz, 5.75 kbit/s), the quality of this coder is better than that of G.723.1 (3.4 kHz, 6.3 kbit/s), and equivalent to G.729 (3.4 kHz, and kbit/s). We also determine the relationship among characterizations of subjective quality, bandwidth and noisiness.