Vocoder

A vocoder, short for voice coder, is a class of speech processing devices or software that analyzes spoken input to extract a compact representation of its spectral and prosodic characteristics and then re-synthesizes speech from that representation, often using a different excitation source. Historically, vocoders were developed to compress voice for transmission and storage by separating spectral information from the excitation signal.

A typical channel vocoder uses a bank of bandpass filters to divide the signal into frequency bands, measures the amplitude envelope of each band, and tracks the fundamental frequency. During synthesis, these envelopes modulate a carrier signal—usually a periodic impulse train or noise—through a synthesis filter bank to produce intelligible speech. This analysis-synthesis framework allows the original voice to be degraded or transformed while retaining recognizable content.

Homer Dudley at Bell Telephone Laboratories built the first practical vocoder in the 1930s, with devices such as the Voder and later the Channel Vocoder. The term vocoder combines elements of “voice” and “encoder/decoder.”

Modern developments encompass several forms: formant vocoders focus on preserving spectral envelope and formant structure; phase vocoders use Fourier analysis for time-scale modification and pitch shifting; and neural vocoders employ neural networks to generate high-quality speech from learned representations. These advances have expanded vocoder use beyond telecommunications to music production, voice creatures, and text-to-speech systems.

Applications include bandwidth-reducing speech transmission, speech synthesis and TTS, and creative voice transformation in music. Limitations of traditional vocoders include artifacts and reduced naturalness at low bitrates, though neural vocoders have significantly improved quality while demanding greater computational resources.