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.

Help on speech recognition system!

Hello! I am creating an speech recognition system with the help of C6713 but i have a little question! I am using the demo aplication dsk_app.c ( located under  /bsl/dsk_app ) and i whant to know: Do i need to overlap the frame for better accuracy? I am now taking a 256 sample frame in a ping buffer, process it and in the mean time the pong buffer is filled. Another question is regarded to the delta and double delta of the cepstrum. Who i can make this operation to be more efficient on this type of application(i mean the edma-ping pong buffers). Best regards, George!

  • Stefan,

    I am moving this post to the C6000 DSP forum, as they will probably have some people with more experience in the C67x device and the EDMA peripheral, ok?

    Despite this, I don't think the accuracy will depend if the buffer is overlapping or you have two ping/pong buffers. The first approach will have a smaller memory usage while the second will guarantee more time to finish processing the buffer.

    Best regards,

    Rafael

  • Hello! there's no problem regarding moving my post, my bad at first. I have another question and i trully hope that i am not getting you bored. I have a 40 filters filter bank(computed similar to those from Sphinx 3) and i've computed the 13 mfcc size vector. Now i what to compute the delta and double delta derivatives. I know that delta = current-2 + current +2. could you give me a hint or something? Thank you very much!  

  • Hi George, 

    I am working with speech recognition now,

    Can you give me your email to contact? 

    Mine is giabaodoan1320@gmail.com

    Best regards!

    Bao