This technology is a software that generates reliable, high-definition timbre spectra for use in speech and voice technologies.
Current software for audio analysis generates timbre spectra using pitch asynchronous methods that fail to separate pitch and timbre completely. With these methods, voice recordings are segmented into frames of 10 to 50 milliseconds and Fourier transformed to yield a timbre spectrum for each frame. When frame size is shifted, timbre spectra become inconsistent because pitch and timbre are always mixed. There is currently no software that separates pitch and timbre completely to yield accurate timbre spectra.
This technology is a software for generating reliable timbre spectra by separating pitch and timbre completely. This software can analyze a recorded signal of speech and generate segments based on whether the sections are voiced or unvoiced. It uses a method of pitch-synchronous segmentation by sensing glottal closing instants to generate pitch periods. Each pitch period is processed via the software’s algorithm and Fourier transformed, yielding accurate, high-definition pitch and timbre spectra.
Patent Pending
IR CU18357
Licensing Contact: Richard Nguyen
