Speech Tool System
This project team will build speech processing tools for experimental speech research in such areas
as speech recognition and speaker authentication.
These tools should include:
A search will be made to find such tools on the Internet
and incorporate them in a directory on Utopia, possibly to be used interactively through a Web interface.
We anticipate finding an appropriate spectral analysis tool.
However, we may have to develop the alignment tool in-house but it is a rather concise algorithm.
- a tool to perform a spectral analysis of an input speech signal
(this usually results in a grey-scale plot of frequency bands as a function of time)
- a tool to segment the speech portion of the signal from the backgraound noise
by threshholding the signal's energy function
- a tool that uses the elastic matching (dynamic time warping) algorithm
to align a speech signal with a pre-segmented one for alignment purposes
Finally, the tools can be tested on the data of Naresh Trilok's dissertation to extend and refine his results.
For example, the authentication phrase "My name is" can be aligned against a presengmented one
that has been divided into the 7 speech sounds so that the system's authentication features
(means and variances) can be automatically extracted.
See Technical Paper
that summarizes the dissertation.