The listening phone

Automatic speech-recognition systems can now support applications from the desktop to public telecoms networks. In comparison with the desktop, the telecommunications network is a challenging environment for speech recognition. The system must work for any speaker and any telephone handset, the bandwidth is restricted and the environment tends to be noisier. Also, feedback, which is restricted to audio responses, is slower than the visual feedback available in a desktop system. Inevitably, telephone-based speech recognition is more restricted than recognition in desktop systems, but this does not prevent many useful applications from being just a phone call away. The author descibes the principles of speech recognition and the systems