Hybrid coding of speech at 4 kbps

We present a novel scheme for hybrid coding of speech signals. This hybrid codec utilizes the excitation/filter model used extensively for speech coding. Similar to other modern vocoders, voiced speech is represented by a frequency domain harmonic model and unvoiced speech by a “noise-like” excitation. However, an analysis-by-synthesis time domain scheme is employed for the transitory portions of the speech signal which cannot be adequately represented by either model. Switching between the time domain and the frequency domain models requires careful handling of the reconstructed linear phase. The structure of a 4 kbps speech codec, based on the hybrid model, is outlined. The new codec shows promise of achieving toll quality at 4 kbps