Delta modulation of pitch, formant, and amplitude signals for the synthesis of voiced speech

A computer simulation was performed to demonstrate the feasibility of delta modulation (DM) as a simple alternative to pulse-code modulation (PCM) for encoding the control signals of a voiced-speech synthesizer. Quantized signals representing the time variations of pitch period, amplitude, and the first three formant frequencies, all band limited to 16 Hz, were available in a 1500-b/s PCM format. Each of the five signals was over sampled at 100 Hz for delta encoding, resulting in a representation at 500 b/s (an information rate utilized recently for an adequate PCM representation of the control signals). Low-pass filtered versions of the DM signals were used to synthesize the original all-voiced utterance with a quality very close to that obtaining in the original 1500-b/s system. Both "linear" and "adaptive" delta modulators were considered; in the latter case, the step size is adapted continuously to the changing slope statistics of an input signal and this provides more efficient encoding. When additional band limiting was applied to the original control signals, resulting in a 700-b/s representation adaptive DM at 250 b/s was sufficient to encode the information without further degradation of the synthetic speech.