how to find phoneme recognition rate from this code ? about deepspeech.torch HOT 5 CLOSED

So when it comes to phoneme error rate I haven't made any modifications to generate phonemes but that may come in the future. That would require altering the way the labels are created and altering the outputs from the neural net.

You can modify the input params (everything is found in AudioData). I use a library written by soumith which provides a simple interface, may be good to check it out!

from deepspeech.torch.

I'll close this for now, but let me know if you make any good progress :)

from deepspeech.torch.

sure i will update you ,if there any progress.

I am doing some experiments with your code to get better accuracy ,For that
i want to what is exactly the output of each frames in spectrogram ie it is
giving word or its character labels [ ie the output of RNN ]

I am doing pre processing the wav files weter it gives a better output or
not.

For the AN4 data base i had got around 49 and 43 accuracy without spell
checker and with spell checker

Any ways thank you for the code U had shared. I am doing my PG in AMRITA
university in coimbathore india.

https://www.*amrita*.edu/

Regards
---Binil---

On Sun, May 1, 2016 at 3:29 AM, Sean Naren [email protected] wrote:

I'll close this for now, but let me know if you make any good progress :)

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#14 (comment)

from deepspeech.torch.

Hey! Check this out, I think you would like this :)

from deepspeech.torch.

thank you very much .

On Wed, May 4, 2016 at 9:17 PM, Sean Naren [email protected] wrote:

Hey! Check this
#18 out, I
think you would like this :)

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#14 (comment)

from deepspeech.torch.