Selective adaptation and utilization of noise reduction technology in call phrase detection

摘要

Techniques are described that selectively adapt and/or selectively employ noise reduction techniques in the detection of one or more features of an audio data frame stream. For example, various techniques relate to selectively adapting and/or using noise reduction techniques, such as detection of a call phrase in an audio data frame stream, detection of a speech characteristic in an audio data frame stream (e.g., for speaker identification), etc. will be. Using the described technique, the characteristics of the audio data frame stream can be more powerfully and accurately detected in various situations, such as environments with strong background noise. In various implementations, the described techniques are implemented in combination with the automated assistant, and the feature(s) detected using the techniques described herein are used to adapt the functionality of the automated assistant.