[This talk is intended to generate discussion, and to solicit feedback for refining a funding proposal I am working on.]
The speech spectrogram is a familiar foundation used in a wide range of speech research tasks - from aiding segmentation to tracking formants to identifying pitch contours. It also has some well-known shortcomings, principle among them the fact that one cannot simultaneously get precise frequency resolution (distinguishing harmonics, or closely-neighbouring formants) and precise time resolution (separating one glottal impulse from the next, for example).
I will present a technique that promise to greatly diminish these problems, known as the "reassigned spectrogram". I will share some existing research, and outline a suite of experiments I plan to conduct to evaluate reassigned spectrograms for two applications: tracking fundamental frequency, and tracking formants. The proposed research will also include developing software to facilitate integration of reassigned spectrograms with other phonetic analysis tasks (specifically, a new module for Praat).