Sampling Rate

The Shannon Theorem or Nyquist Law require that the sampling rate is higher than twice the maximum frequency in the digitized signal. Since speech is more or less located below 8kHz most speech corpora have a sampling rate of 16kHz minimum. Exceptions are telephone recording where the bandwidth is technically reduced to 300Hz - 3300Hz and usually a sampling rate of 8kHz is used. Since the audio CD standard was introduced with 44,1kHz also the dividers 22,05 kHz and 11,025 kHz are used because some audio devices do not process other sampling rates than these. This is also the reason why we recommend avoiding `exotic' sampling rates, whenever possible.
Laryngograph signals are usually sampled at the same frequency as the speech signal. Because of their low bandwidth speech movement data (e.g. EMA^4.5) can be sampled at about 200 Hz.

• Telephone Recording: 8kHz
• -site or field recording: 16kHz, 22,05kHz
• signal: minimum 16kHz
• EMA signals: 200Hz