Here we assume each second of audio is divided into 50 spectrogram frames, and an individual’s speech in any given spectrogram frame consists of one of 10,000 possible sounds. The n-talker model must ...