The European Telecommunications Standards Institute (ETSI) has approved a speech recognition algorithm developed by Motorola Labs, France Telecom (FT) and Alcatel. The feature extraction- based algorithm was competing with another developed by Qualcomm for ETSI approval.
John Horrocks, chairman of ETSI's speech transmission and quality technical committee, says the institute had drawn up a list of criteria to define recognition accuracy in order to rank the two algorithms.
The average recognition accuracy of the latest algorithm was more than 90% in a noisy car environment. Five languages were used in the testing. This is a 53% performance improvement on the previous generation standard.
Horrocks said: "The distributed speech recognition algorithm processes speech at microphone quality to extract key features which are sent across a poor quality link."
The use of a low-standard link makes it applicable for devices such as mobile phones, where speech has often to compete with a lot of noise.
ETSI plans to publish the algorithm in two or three months' time. Motorola, FT and Alcatel will then licence it to device manufacturers.