Formant Frequency

Formant frequencies, the resonant frequencies of the vocal tract, are crucial acoustic features in speech, informing both speech perception and production. Current research focuses on improving the accuracy and robustness of formant estimation and synthesis, employing techniques like neural networks (including diffusion models and source-filter models with differentiable resonant filters) and wavelet transforms to achieve this. These advancements are driving improvements in speech synthesis, particularly in achieving higher quality and more natural-sounding speech, and are also contributing to a deeper understanding of vocal tract dynamics across different speakers, ages, and time periods.