(Extremely) Simplified Model of Speech Production Periodic source Noise source Filters voiced unvoiced Coupling Speech.

(Extremely) Simplified Model of Speech

Production

Periodicsource

Noisesource

Filtersvoiced

unvoiced Coupling Speech

Channel Vocoder, Synthesizer component

Channel Vocoder, Analyzer component

Vocoders: vocal tract synthesis

•Short time spectral synthesis

•Linear predictive synthesis

•Cepstral synthesis

•Formant synthesis

Vocoders: vocal tract analysis

•Short time spectral analysis

•Linear predictive analysis

•Cepstral analysis

•Formant analysis

Vocoders: vocal source analysis

•Pitch detection

•Voiced/unvoiced decision

Fine structure and Spectral Envelope

Narrow and wideband filterbank spectral analyses

Wideband spectrogram

Figure 3.5: Narrow band spectrogramNarrowband spectrogram

•Source-Filter Separation by Cepstral Analysis

LPC spectral envelope estimation

Formant Analysis•Peak picking from smoothed spectrum

• Root-finding from LPC polynomial - finding second-order sections

•In general, position more important than bandwidth

•Deterministic or statistical analysis to find best formant “track”

http://ptolemy.eecs.berkeley.edu/~eal/audio/vocoder.html

Ed Lee Vocoder Page

http://www.sirlab.de/linux/descr_vocoder.html

Linux Vocoder Page

How does the speech sound with the middle spectrum differ from the left one?

From the right?