Disclosed are methods and systems for suppressing hot words, and systems including computer programs encoded on a computer storage medium.In one aspect, the method includes the action of receiving audio data corresponding to the reproduction of the vocalization.Action is(I) configured to determine if a given audio data sample contains an audio watermark(II) each includes an audio watermark sampleAudio data samples with watermarksAnd each audio watermark sample does not includeA model trained using watermark free audio data samplesFurther include giving audio data as input.The action further includes receiving data indicating whether the audio data contains audio watermarks from the model.The action further includes determining whether or not audio data is processed or stopped based on the data indicating whether the audio data contains audio watermarks.
展开▼