Speech coders Types:
1. waveform coders:
speech modeling is not required
-DPCM: transmit difference between successive sample values, d[k] = m[k] – m[k-1], and reconstructed by adding d[k] to previous sample m[k-1]
2. Source coders
modeling vocal tract, vocal tract filter parameters are created at encoder, model parameters are sent to receiver, and decoder will reconstruct speech based on received parameters, high complexity, better compression ratio, quality of speech is low
– LPC, modeled using a discrete time linear phase filter, constructed from all-pole transfer function H(z), transmit the selected characteristic of error signal like gain factor, pitch information, voice/un-voice decision information.
multi-pulse excited LPC, regular pulse excited, code excited, vector sum excited, residual excited, mixed excited
– channel vocoder, bank of bandpass filter
– formant vocoder, code only most important pulse, formant(peak)
– cepstrum vocoder, separate excitation and vocal tract coefficient and send only vocal tract coefficient.
– voice excited vocoder, uses narrow band LPF