How do you control a radio using speech recognition when it’s playing in the background? Over the last few years speech recognisers have been designed to cope with relatively steady background noises such as car or aircraft noise, but sounds such as music or other speech can upset them.
At the <b>IEEE workshop</b> on “”;Robust Methods for Speech Recognition in Adverse Conditions””; held this week in Tampere, Finland, Michael Carey, Ensigma’s Managing Director, presented a new algorithm for the cancellation of interference from in-car entertainment and other similar stereo sources.
The patented algorithm is specifically designed for use with speech recognisers unlike the adaptive filters usually used. It operates by estimating the spectrum of the in-car entertainment signal and subtracting this from the microphone signal before it is input into the recogniser.
Carey commented, “”;We are pleased that the algorithm performs well on a stereo signal as that is not an easy problem. Also it requires very little computation, less than 1 MIP on a typical DSP. We will now add it to our Robust Recogniser which already works well in a car travelling at 120 kph.””;
The Signal Spectrum Subtraction algorithm will be incorporated into Ensigma’s Speech Recogniser and Ensigma is actively seeking partnerships with handset, automotive electronics, telecommunication and semiconductor manufacturers to take it through to realisation.