Abstract: A service platform for the provision of automated speech based services over a telephone network comprises a speech processor including associated memory, to provide at least one of speech recognition, speaker recognition, speaker verification, speech synthesis; line interface means for terminating a number of telephony channels of the network; a speech bus to carry speech signals between the line interface means and the speech processor; a control processor to control the routing of speech signals over the speech bus; a control bus connecting the control means with the speech processor and the line interface; a separate memory holding speech processing software; and a data link connecting the memory means with the speech processing means and operable to load programs into the speech processor during rum time in the event that the appropriate software does not exist on the speech processor in a free state.

Abstract: In a speech processor such as a speech recognizer, the problem of distortion of extracted features caused by adaptation of the input automatic gain control (AGC) during feature extraction is solved by storing the AGC's gain coefficient along with the energy level of each extracted feature. At the end of the sampling period the stored gain coefficients are set equal to the minimum stored coefficient and the associated energy levels adjusted accordingly. The AGC circuit may comprise a digitally switched attenuator under the control of a microprocessor performing the speech recognition.

Abstract: Residual noise control is provided on telecommunications links which are at least partly 4-wire, by means of a variable threshold center clipper 30 applied to the output of an echo canceller 100. The threshold of the center clipper 30 is varied in accordance with the magnitude of an echo estimate 21 generated by the echo canceller 100.In order to prevent mutilation of near end speech the clipper is disabled when such speech is detected. However, the echo canceller 100 remains in circuit to allow control of echo even during doubletalk.The clippers threshold may be controlled in response to the instantaneous modulus of the echo estimate. Alternatively, the echo estimate may be fed to a modular 60 and filter 55 to provide a short term filtered version of the echo estimate which allows control of other components of residual noise.