1. Draw a block diagram illustrating the key stages of an A-by-S based speech coder. Briefly explain each block.
Excitation Generator: Determines the excitation signal to use in the speech synthesis stage.
Inverse Long Term Predictor: Processes the excitation to produce a Linear Prediction Residual Signal
Inverse Short Term Predictor: Applies an LPC synthesis filter to the residual signal to produce synthesised speech.
Summing stage: Subtracts the original speech from the synthesised version to produce an error signal.
Error minimisation: Determines an parameters based on the error signal to be used to adaptively choose an alternative excitation signal such that a new version of the synthesised speech has a lower error than the previous version.
2. Draw a block diagram of a CELP speech coder. Briefly describe each block.
Adaptive Codebook: This codebook is used to quantise the residual signal. It replaces the LTP filter that is used in other speech coders. It stores a series of delayed segments of the previous excitation.
Adaptive Gain: This is used to amplitude scale the chosen code vector from the adaptive codebook, such that the overall average magnitude (gain) matches the gain of the current residual being quantised.
Fixed Codebook: This codebook is used to quantise the excitation. It consists of code vectors, each representing a possible excitation waveform for a segment of speech. During coding, each code vector is tested and the one which produces the smallest error between the current original and synthesised speech segment is chosen and the corresponding index is transmitted.
Fixed Codebook gain: Similar to the ACB gain, this is used to amplitude scale the chosen code vector from the fixed codebook, such that the overall average magnitude (gain) matches the gain of the current residual being quantised.
Weighted short term synthesis filter: This is used to perform LPC based synthesis of speech using the residual obtained from the combination of the ACB and FCB code vectors and gains.
Error minimisation: This is used to test every code vector in the ACB and FCB to find the ones that lead to the lowest error between the original and synthesised speech signals.
Weighting filter: This is used to modify the speech signal such that those frequency components of perceptual significance are emphasised and those of less perceptual importance are de-emphasised. In the frequency domain, this would equate to an increase (for emphasis) and decrease (for de-emphasis) of the magnitudes at different frequencies.
Zero response: This is required due to the truncation of the impulse response of the IIR LPC synthesis filter to allow synthesis using finite length signal segments. As a result of this truncation, if
