Extracting phonatory features from the running speech.
First, a voice-activity-detector (VAD) is applied to the speech signal, followed by a voicing check. In both cases, the PRAAT tool is used. If a voiced segment is longer than 100 ms, the phonatory features are extracted. Otherwise, the voiced signal is not long enough to obtain the features such as jitter or shimmer, as not enough vocal tract cycles are repeated. The fundamental frequency for someone may be less than 100 Hz, corresponding to 10 ms, and at least five cycles are needed to obtain these features.
- the main function is the file: get_features_running.m
Supported by the Quality Internal Grants of BUT (project KInG, reg. no. CZ.02.2.69/0.0/0.0/19_073/0016948; financed from OP VVV).