An LPC vocoder takes a speech waveform sampled at 8kHz and compresses it to a lower bitrate for transmission by modeling the human vocal tract as a linear system over frames of 25ms. It estimates the vocal tract spectrum and pitch in each frame, transmitting this data which is much smaller than sending the raw samples. At decoding, the estimated spectrum is excited with an impulse train of the estimated pitch to reproduce the speech. The document also describes a gain estimation method that utilizes speech waveform envelopes to estimate gains for voiced and unvoiced frames such that the synthetic speech amplitude matches the envelope.