Processing group delay spectrograms for study of formant and harmonic contours in speech signals

Area of Science:

• Speech Processing
• Acoustic Phonetics
• Signal Analysis

Background:

• Understanding speech signal characteristics like formants and harmonics is crucial for speech analysis and synthesis.
• Traditional methods for analyzing these features can be complex and may require phase unwrapping.
• Group delay (GD) spectrograms offer a potential alternative for visualizing spectral information without phase wrapping.

Purpose of the Study:

• To investigate the utility of group delay (GD) spectrograms for observing formant and harmonic contours in speech signals.
• To evaluate the accuracy of GD spectrograms in representing known formant contours for synthetic and natural speech.

Main Methods:

• Processing of group delay (GD) spectrograms derived from speech signals.
• Utilizing modified single frequency filtering (SFF) analysis to generate instantaneous wideband and narrowband equivalent GD spectrograms.
• Observing formant and harmonic frequency contours by identifying peaks in the respective GD spectrograms.

Main Results:

• Formant frequency contours were successfully observed in the GD spectrograms.
• Observed formant contours closely matched ground truth contours for synthetic speech signals.
• Observed formant contours showed approximate agreement with ground truth for natural speech, particularly in voiced regions.

Conclusions:

• Group delay spectrograms provide a viable method for visualizing formant and harmonic contours in speech.
• The modified SFF analysis effectively derives GD spectrograms suitable for contour observation.
• Further research is needed for automated formant frequency extraction, focusing on spurious point elimination.