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.

MSP432P401R: Noise Cancellation in MSP Speech Recognizer

Part Number: MSP432P401R

Hi,

I have used MSP Speech Recognizer and it works very well in an office environment. When there is any kind of noise along with the speech, it doesnt detect the keyword spoken.

Is there any way around for this?

I am trying to implement a filter to remove the noise from data stored in audio buffer, after which its sent for recognition.

Do you think this will work? If not, please suggest.

By Noise I mean to include- sound of wind, music , or people talking at a distance.

  • Hi Abhinandan,

    We have tested the speech recognizer in various noise environments. Indeed the recognition becomes more difficult in noisy conditions. As with most recognizers, performance is a function of the signal to noise ratio. The performance tends to start degrading as the SNR drops below about 15-20dB. The conditions you mention are challenging, especially music and other people talking over the user's speech, which distort the speech signal in unpredictable ways.

    The speech recognizer does contain some algorithms to aid in maintaining performance in noisy environments. Incoming audio data is filtered with a simple high-pass filter to remove low frequency energy. Such energy is often found in automobile and machinery noise. Other algorithms attempt to compensate for low to moderate spectrally stationary noise. Implementing a filter in an attempt to remove noise may help, but the challenge is to maintain the undistorted spectral content of the speech at the same time. For situations like interfering music and speech, using a simple filter will not likely be effective.

    More sophisticated speech recognizer systems, such as the currently popular voice assistants, utilize multiple microphone arrays with beamforming, and complex noise cancellation and source separation algorithms to obtain higher SNR signals. But this requires much additional processing, and hence power. The goal of the MSP speech recognizer is to develop a low-power single microphone system, which has its limitations.

    I am currently working on some modifications for the next release that should help improve performance in somewhat lower SNR situations. However, it does not address the difficult problems like removing interfering music and speech.

    Regards,
  • Hi Lorin,

    Thanks for reply. Looking forward for the new version release.
    Anyways I guess I'll have move on to test the particular scenarios I mentioned on a DSP board ( assuming that MSP doesnt has the required capabilities). Any suggestions on which board might be suitable for the mentioned scenarios.
    Also I do have a C5535 Eval board which supports TI ESR . Will that work ?

**Attention** This is a public forum