Vocoding Techniques

Speech Sounds

Speech sounds are generated when air from the lungs flows through the vocal tract, which includes the mouth, pharynx, throat, and nasal cavities. If the vocal folds are tightened appropriately, they will start oscillating in the airflow, creating periodic waveforms known as voiced sounds. The frequency of this oscillation is perceived as the pitch of the sound. Should the vocal tract have constrictions, such as when the lower lip is pushed lightly against the upper teeth, airflow is partially obstructed causing turbulences, which are recognized as unvoiced sounds. These two sounds, periodic oscillations and noiselike turbulences, are known as the speech source, and they are physiological effects, which can be colloquially described as the buzz and hiss of speech sounds, respectively.

[Page 2084]Another important component of the speech sound is the shape of the vocal tract, which influences the timbre of the speech sound and is determined mainly by the position of the tongue, lips, and jaw. In fact, different configurations of the vocal tract shape give rise to acoustic resonances, which are recognized and identified as different vowels. The shape of the vocal tract has therefore a direct connection with the uttered phonemes. That is, the meaning of the words spoken can change depending on the vocal tract shape.

How Vocoders Work

Vocoders analyze and subsequently modify these characteristics of sound signals. By synthesizing sounds that share the characteristics of the speech source and resonances of the vocal tract, sounds that perceptually resemble speech can be obtained. In particular, when a sound shares the resonances and acoustic shaping effect of the vocal tract, it will sound like a vowel, even if the sound source would not be related to a speech signal.

In telecommunication, all mainstream standards for digital transmission of speech (e.g., standards associated with mobile phones) are based on the vocoding principle. At the transmitter, speech signals are analyzed for their pitch and turbulent noise as well as acoustic shaping due to the vocal tract, and this physiologically motivated description of the sound is transmitted in a digital form. The receiver can then reconstruct a speech signal, which sounds similar to the original sound. As long as the regenerated speech sound shares these basic characteristics of the sound source and vocal tract obtained from the original speech signal, it will be intelligible and sound like the original, even if some other features of the signal would be reproduced with less accuracy. Use of a vocoder thus gives a compact representation of the speech signal, such that the amount of data transmitted is reduced, typically from 96 kbit/s of the original signal down to 8 kbit/s of the transmitted signal. Such compression of speech signals reduces costs of telecommunication by reducing energy consumption, bandwidth, and infrastructure requirements.

...