Speech and Audio Processing

Lecture-9 By: Mohit Goel

LINEAR PREDICTIVE CODING

LINEAR PREDICTIVE CODING

The white noise signal x[n] is filtered by the AR process synthesizer to obtain s[n]the AR signalwith the AR parameters denoted by A linear predictor is used to predict s[n] based on the M past samples.

LINEAR PREDICTIVE CODING

This is done with

where the ai are the estimates of the AR parameters and are referred to as the linear prediction coefficients (LPCs). The constant M is known as the prediction order. The prediction error is equal to

LINEAR PREDICTIVE CODING

Error Minimization
The mean-squared prediction error is minimized by selecting the appropriate LPCs.

The optimal LPCs can be found by setting the partial derivatives of J with respect to ai to zero; that is,

Error Minimization
Error will be minimum when

Error Minimization

Above equation allows the finding of the LPCs if the autocorrelation values of s[n] are known from l = 0 to M.

Prediction Gain
The prediction gain of a predictor is given by

and is the ratio between the variance of the input signal and the variance of the prediction error in decibels (dB).

Minimum Mean-Squared Prediction Error

When that is, the prediction error is the same as the white noise used to generate the AR signal s[n].

Minimum Mean-Squared Prediction Error

Indeed, this is the optimal situation where the meansquared error is minimized or the prediction gain is maximized., with The optimal condition can be reached when the order of the predictor is equal to or higher than the order of the AR process synthesizer. A simple method to estimate M from a signal source is by plotting the prediction gain as a function of the prediction order. In this way it is possible to determine the prediction order for which the gain saturates; that is, further increasing the prediction order from a certain critical point will not provide additional gain.

LINEAR PREDICTION ANALYSIS OF NONSTATIONARY SIGNALS

Prediction Schemes Two main techniques are applied in speech coding: internal prediction and external prediction. Figure drawn bellow illustrates the schemes

Prediction Schemes
For internal prediction, the LPCs derived from the estimated autocorrelation values using the frames data, are applied to process the frames data themselves. In external prediction, however, the derived LPCs are used in a future frame. The reason why external prediction can be used is because the signal statistics change slowly with time. If the frame is not excessively long, its properties can be derived from the not so distant past.

Prediction Schemes
A longer frame has the advantage of less computational complexity and lower bit-rate, since calculation and transmission of LPCs are done less frequently. However, a longer coding delay results from the fact that the system has to wait longer for sample collection. Also, due to the changing nature of a nonstationary environment, the LPCs derived from a long frame might not be able to produce good prediction gain. On the other hand, a shorter frame requires more frequent update of the LPCs, resulting in a more accurate representation of the signal statistics. Drawbacks include higher computational load and bit-rate.

Prediction Gain

The LPCs are found from the samples inside the interval [m - N + 1;m for internal prediction, and n < m - N + 1 for external prediction.