(i) The performance of the multi-pulse linear prediction based vocoder can be further improved by using the vector quantization technique for coding of the excitation signal.
(ii) The technique is known as Code Excited Linear Prediction (CELP).
(iii) This is because the excitation signal is taken from a code book.
(iv) The speech synthesizer in CELP-based vocoder is identical to the one used in adaptive coder.
(v) Examination of a large number of speech signals has indicated that the excitation sequence for LP vocoder use Gaussian in nature.
(vi) Hence, a basic CELP coder uses a Gaussian code book.
(vii) The error is perceptually weighted, the mean squared error is computed and it is then minimized.
(viii) The minimized perceptual error is used to change the code word from the code book. This process continues until the mean square error is minimum.
Modifications to basic CELP
(i) Another possible choice for code book entries is center-clipped Gaussian vectors.
(ii) This choice reduces the complexity of the excitation vector and improve the performance.
(iii) The center-clipped waveform will preserve the excitation pulses.
(iv) The Self-Excited Linear Prediction (SELP) is an improvement the CELP coder this system uses the excitation signals derived from the past history of the coded excitation itself.
(v) Initially the memory is filled with random entries.
(vi) For each coded vector, the sequence with some index is selected for optimal coding. The best vector selected is passed to the code book and the oldest entry is deleted. Hence, SELP is CELP with an adaptive code book.
(vii) An initial pitch calculation is done on perceptually weighted original speech to get the idea of pitch value.
(viii) A small update is done by the pitch prediction process.
(ix) This reduces the computation time for pitch prediction.
(x) As the pitch value changes by 1%/ms, the pitch updates more frequently.
and 3 others joined a min ago.
teamques10
★ 67k