Automatic conditioning of speech signals

As a part of the larger problem of providing speech input to a computer in a communications system, before either digital buffering, vocoding, or programmed recognition of the speech signals is performed, automatic conditioning of the input signal can significantly improve system performance. The signal conditioning sub-system described in this paper is intended to provide a constant-amplitude signal of the best quality possible from an uncontrolled speech source such as a telephone line, and to indicate the beginning and continued presence of a speech utterance. A fast-acting wide-range automatic level control is used both to compensate for the average strength of the received signal and to remove the word and phrase amplitude dynamics so that the amplitude-controlled signal is contained in a minimum dynamic range. The presence of an utterance is detected by comparison of the logarithm of signal amplitude to an automatic variable threshold set by the logarithm of the input noise as measured during the absence of signal, thus achieving a constant SNR threshold. The functions described for conditioning can be implemented with analog circuits without the aid of signal delay.