Espy-Wilson Awarded NSF Grant, Assists with Navy STTR Grant
NSF: Multipitch Tracking
The one-year, $125,000 NSF award is for Robust Intelligence: Extension of the APP Detector for Multipitch Tracking and Speaker Separation.
Because of the loss of fine temporal structure, users of cochlear implants encounter problems in separating speakers in multi-speaker environments. Speech recognition and speaker identification systems must deal with simultaneous speech from several talkers in conversations. A crucial preprocessing step for such systems is segregating speech according to its constituent sources.
Because of the differences in people's vocal chords, different speakers have characteristic pitch ranges. Pitch tracks can be used to help separate the combined signal into different speech streams. The research will help systems recognize the number of speakers and the separation of their pitch tracks based on the periodic portions of the speech signal (i.e., voiced regions).
Current popular multi-pitch tracking approaches are susceptible to artifacts caused by the interaction between the periodic regions of the different speech signals. Consequently, the periodicity of the combined signal can be different from that of the individual components. The major new idea is the extension of an existent periodicity and pitch estimation process to higher dimensions, arriving at a multi-dimensional periodicity function which is not susceptible to the harmonic interaction artifacts.
The multiple pitch tracks obtained are more accurate even when one speaker is considerably more dominant than the other. The approach is easily generalized to non-speech audio and should be robust in noisy channels. This project is a step towards separating actual speech streams from each other based on multi-pitch information.
Navy STTR: Speech Separation
Espy-Wilson is also a subcontractor for a Navy Phase II Small Business Technology Transfer (STTR) award to Signal Processing, Inc. of Rockville, Md., A Novel Speech Separation Approach for Enhanced Speaker Identification and Speech Recognition. This is an 18-month award for $103,066.
In this grant, Signal Processing, Inc. will use a speech enhancement algorithm developed by Espy-Wilson in a noise suppression system it is developing. Espy-Wilson’s algorithm is based on modified phase opponency (MPO) and does not require noise estimation.
Once developed, the company’s system could suppress high background noise and achieve high performance speech separation, speaker identification, and speech recognition. It could also be used for bird monitoring to avoid birdstrikes to aircraft, and speech enhancement in communication centers, conference rooms, aircraft cockpits, cars and buses. Further uses could be in security monitoring in airport terminals, bus and train stations, where the system could pick up multiple conversations from different people and different angles and as a front-end processor to an automatic speech recognition system.
September 14, 2008