LPC Vocoder Final presentation Hongxu Zhao Shiqi Li
Outline Revisiting the topic Implementation & Demonstration
Revisiting the topic
LPC vocoder LPC – Linear Predictive Coding Block diagram
Encoder Speech is segmented into 10 -30 ms intervals Coefficients for an M-order model are determined to minimize the prediction error for a linear filter over the interval The type of speech is determined - voiced, unvoiced For voiced speech, pitch period is determined Gain is estimated
Decoder The encoded gain, LP coefficients, speech type and pitch period are decoded Depending on the speech type, the excitation source is chosen as either white noise or a periodic impulse train The excitation source is scaled according to the transmitted gain and passed through an inverse of the analysis filter to produce synthesis speech
Implementation & Demonstration
Different implementations V/UV detection Energy Zero-crossing rate Pitch detection AMDF Autocorrelation LPC order 10 for voiced, 4 for unvoiced 12 for both Sampling frequency 16 KHz & 8 KHz Windowing, overlapping