Analytical speech encoding

Digital encoding of speech with compression by means of the CELP parametric technique is now well established. Toll quality transmission can be achieved at rates down to 8 kb/s. The paper describes a new waveform approach to speech coding. The advantages of waveform coding when compared to parametric coding are, independence with respect to the signal nature, a performance measure by means of the signal to encoding noise ratio, and simpler algorithms leading to a digital implementation at low cost with small encoding delay and small power consumption. It is well known, from analytical signal theory, that a band limited message m(t) can be written m(t)=A(t)cosφ(t) where, with mˆ(t) is the Hilbert transform of m(t), A(t) and φ(t) are the amplitude and phase of the analytical signal m(t)+jmˆ(t). A(t) and φ(t) can be found either by taking the inverse Fourier transform on the positive frequencies of m(t) or by single sideband modulation followed by envelope and frequency detection. A preliminary estimate of the bandwidth of A(t) and φ(t) can be found by assuming a sinusoidal modulation of the instantaneous phase. An upper bound follows from Carson´s rule F<B/4 where B is the bandwidth of the message. For telephone speech F<850 Hz. Experimental results show that the bandwidth is actually below this estimate. Therefore encoding at 8 kb/s is quite feasible