This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

MSP432 Speech recognition in a noisy room

Hello MSP team,

Do we have any information on speech recognition in (noisy) rooms. e.g.: howto pick up a voice somewhere in the room with one microphone and be able to recognize speech. Do we have implementation examples using AGC and or noise cancellation?

Thanks in advance, Patrick Verkennis

  • You may want to start with this.

  • Thanks, I have had a look a this and this is a very good starting point.

    Again, which techniques can be used to have room coverage with a single fixed microphone. (things like AGC, noise cancellation)
  • Speech recognition in a noisy environment with a single microphone is indeed challenging. Performance is tied closely to the SNR of the speech signal. AGC does not do anything to change the SNR itself, so is of limited use. For noise cancellation to be effective, it must not cancel any of the speech signal, which is difficult to determine.

    There are two typical approaches taken in speech recognition. One is to measure the background audio signal, and compensate the speech recognition models assuming that the background signal statistics remain constant over the short duration of an utterance. The second method is to estimate the predicted clean speech features given the statistics of the noise and input the clean speech features to the recognizer. The speech recognizer implemented on the MSP432 uses the former approach.

**Attention** This is a public forum