Real-time pitch tracking by detection of glottal excitation epochs in speech signal using Hilbert envelope

A system for real-time pitch tracking by detection of glottal excitation epochs in a speech signal, the system comprising:

• a dynamic range compression module configured to apply a dynamic range compression on the speech signal to obtain a dynamic range compressed signal;
• a Hilbert envelope calculation module configured to calculate a Hilbert envelope of the dynamic range compressed signal;
• and
• an epoch marking and pitch detection module to process the Hilbert envelope for marking epochs and outputting pitch periods, wherein the epoch marking and pitch detection module comprises:
• a dynamic peak detector module configured to obtain a peak envelope from the Hilbert envelope;
• a nonlinear smoother module configured to calculate a smoothed peak envelope from the peak envelope;
• and
• a saliency detector module configured to mark epochs and output pitch periods from the smoothed peak envelope by detection of a saliency,
• wherein the saliency detector module comprises:
• a saliency enhancer module configured to obtain a saliency-enhanced peak envelope by processing the smoothed peak envelope to emphasize the points with high rate of change;
• an amplitude-duration thresholding module configured to apply amplitude-duration thresholding on the saliency-enhanced peak envelope, with an amplitude threshold and a duration threshold, to mark the epochs and output the pitch periods;
• an amplitude threshold calculator configured to calculate the amplitude threshold as a short-time average magnitude of the saliency-enhanced peak envelope;
• and
• a duration threshold calculator configured to calculate the duration threshold as half of the mean of the preceding ten pitch periods which are lying within a set range of 2-15 ms and apply a lower limit of 2 ms.