• State Resolved
  • Locked Locked
  • Replies 6 replies
  • Answers 2 answers
  • Subscribers 98 subscribers
  • Views 2042 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

TMDX5535EZDSP: Active/Adaptive Noise Cancellation support in speech recognition

Abhinandan BR
Intellectual 945 points
Part Number: TMDX5535EZDSP


Hi,

I am working on speech recognition. My requirement is to be able to filter out unwanted background speech or music (usually low in amplitude) and consider only high amplitude voice data for recognition. 

I am using a C5535 EZDSP board. I am using TI ESR library for speech recognition on the board. Although it works well, it does not have adaptive/active noise cancellation out of the box. I am planning to integrate it with a mic array and perform filtering of unwanted speech data. Is this possible on this platform?

Is there any add on board which can filter out unwanted speech, available as a separate board? or which can be integrated with c5535?

Also please suggest other TI boards if any support my requirement .

PS: By Noise cancellation, I mean filtering of unwanted sounds/speech/music.

  • 0
    TI__Guru 65405 points
    Hi Abhinandan,

    I've forwarded this to the C55x SW experts. Their feedback should be posted here.

    BR
    Tsvetolin Shulev
  • 0
    TI__Mastermind 30365 points
    Abhinandan,

    Please take a look at www.ti.com/.../TIDEP-0077 this is a TI design on the C5517. Here is a demo video as well www.youtube.com/watch

    This can be ported to the C5535 but there are several considerations such as MIPS, memory, number of I2S lines, etc.

    Lali
  • 0 in reply to Lalindra Jayatilleke
    Intellectual 945 points

    Hi Lali,

    Thanks for the suggestion.
    I would like to combine Audio Pre-processing with TI ESR demo project.
    Is it possible to achieve that in C55xx series DSP ? If yes, I figured that the memory ( 320KB of RAM memory) left out after audio pre processing wont be enough to run TI ESR demo. Correct me if I'm wrong in this.

    If my memory analysis is right. Could you suggest an alternative for this? 

    I think an add-on board which can perform all the necessary pre-processing before the audio is sent to c5535 as input, will solve my purpose, if doing both pre-processing and recognition is not possible on c5535.

  • 0 in reply to Abhinandan BR
    TI__Mastermind 30365 points
    Abhinandan,

    It might be possible, but this has not been tried on the C5535 with TIesr + Voice processing. We typically request customers to work with the Sensory solution which is a production quality voice triggering solution that is far better than TIesr and works on the C5000 family of devices.

    You can certainly try to integrate TIesr into the C5517 with voice processing, but just to set expectations, support is limited for this combo.
    If you would like to work with the Sensory solution, please contact a local TI representative or Sensory directly to get started.

    You could also think along the lines of 2 x C5535s if entertaining an addon board.

    Hope this helps.

    Lali
  • 0 in reply to Lalindra Jayatilleke
    Intellectual 945 points

    Hi Lali,

    Thanks for suggestion. I do not want to go for Sensory as of now.

    Before I start with other option, I want to know if pre-processing can remove both speech and background noise, to give me output of clear voice. If any accuracy or test reports available please share. 

  • 0 in reply to Abhinandan BR
    TI__Mastermind 30365 points
    Abhinandan,

    Please take a look at this video for a demo of the TI design www.youtube.com/watch
    We don't yet have any specific test results on the system. The key algorithms in the software are a Beam Former (BF) & Adaptive Spectral Noise Reduction (ASNR) .

    The intention of BF is steer a beam in a specific direction. Here's a good reference www.labbookpages.co.uk/audio/beamforming/delaySum.html

    The ASNR is great for some of the low frequency noise. Here is a good white paper further explaining.
    software-dl.ti.com/.../BF_white_paper.pdf

    Lali