Computer Speech Technology

Computer Speech Technology

Cambridge Algorithmica has worked extensively on Computer Speech Technology (CST), including:

Automatic Speech Recognition (ASR)
Speaker Verification (SV)
Speaker Identification (SID)
Speech Activity Detection (SAD)
Text-to-Speech Synthesis (TtS)
Speech Coding: waveform coders and vocoders

On ASR, work has included:

Extensive applied research on Large Vocabulary Automatic Speech Recognition (LVASR), using novel techniques such as variable-duration segmentation for acoustic-phonetic transcription.  The most recent research offers improved performance using a new approach that requires multi-processor computer power, of the sort that is now becoming available on up-market desktop computers.

Investigation of improvements to ASR in military fighter aircraft, where speakers use a breathing mask (which degrades the speech signal) and are subject to positive pressure breathing (which causes further degradation by inflating the vocal tract).

A prototype telephone banking system using ASR.  This involved dialogue design, including full recovery from dialogue errors.  The design was ported for real-time evaluation over the public telephone system, using a commercially available ASR engine and a bespoke simulation of the back-end banking system.

The Company's work on modelling of Probability Density Functions (PDFs) is also highly relevant to work on ASR, SV and SID.  This work is covered under biometrics (particularly multi-biometric fusion).  Click on this crosslink for some of Cambridge Algorithmica's publications on biometrics.



Home | About | Speech | Modems and Protocols | Biometrics | Condition Monitoring | Publications | Contact