Linear Predictive Coding (LPC) is a transformation method used in speech compression of mobile radio systems. The method works with filtering techniques and filters out speech-relevant components such as the fundamental speech frequency, loudness
and voicing from a mixture of signals. For this purpose, the digital speech signal is transformed into an excitation signal and into coefficients of a time-varying filter. This is done by regular sampling (160 samples per second) of the speech signal to be transmitted, from which filter coefficients are then formed and transmitted to the decoder. When the signal is divided into short-duration segments, it is assumed that the change between two segments is marginal. Therefore, by using linear prediction, only the changes between two consecutive short-time segments need to be transmitted, which can significantly reduce the file size
.With LPC coding, data
rates of6 kbit/s to 24 kbit/s can be achieved at sampling rates
and 16 kHz. Themethod is also used for voice coding in GSM
, among other applications.
The difference between Linear Predictive Coding (LPC) and Adaptive Predictive Coding (APC) is that the APC method uses fewer prediction coefficients and therefore offers a higher sampling rate than LPC coding.