Increase the accuracy of the converted text from audio file

[Karthika]

Hi,

I am currently using the "Pocketsphinx" to convert the speech to text (in offline for saved audio files).
As I have used and tried many options,still I am getting accuracy as 50% only and the words are not silmilar as spoken in the audio.

Tried in another way by "Adaption Process" also ,still showing the same percentage.

I am not sure how to increase the accuracy as same as in the audio.

Please give your suggestion on how to increase the "Accuracy" in pocketsphinx.

Thanks in advance.

[Nickolay V. Shmyrev]

To get help on that issue you need to provide the test data you are trying to decode an other models you used to estimate the accuracy.

[Karthika]

Thanks for your reply.

Yes Nickolay I already tried those different models((i.e)Acoustic and Language Models) that metioned in this link https://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/US%20English/ and comparing with this different model only finalised the accuracy of 50% in this model cmusphinx-en-us-8khz-5.2.tar.gz.

And one more thing for some files it is showing the accuracy like this :

Words: 238 Correct: 51 Errors: 356 Percent correct = 21.43% Error = 149.58% Accuracy = -49.58%
Insertions: 169 Deletions: 0 Substitutions: 187
TOTAL Words: 238 Correct: 51 Errors: 356
TOTAL Percent correct = 21.43% Error = 149.58% Accuracy = -49.58%
TOTAL Insertions: 169 Deletions: 0 Substitutions: 187

This is showing in negative,because the words that is spoken is not returning in the text file,that is the problem.

Please help on this and give your valuable suggestions to get a correct data as of in the Audio file.

Thanks in advance

[Nickolay V. Shmyrev]

These results suggest your data is not in a proper format. You can consult documentation for details.

[Karthika]

Ok Thanks,let me check that.

But already cheked that properties and options.

Will you please provide me the information of how to give the proper date.

It will be really helpful for me.

I am tried those suggested in the below link also.

http://cmusphinx.sourceforge.net/wiki/tutorialtuning

Thanks in advance.

[Karthika]

Hi Nickolay,

Sorry for bothering you.

I have refererred some of your blogs and find that Kaldi is somehow better interms of accuracy,comparing to CMUsphinx.

Here is the question,As I am new in speech recognition and not able to understand all the stuffs regarding the accuracy.So which will be better for the person who is like me to start the speech recognition development.(i.e) Kaldi or CMUSphinx ??

Please give your suggestion on this,as I am spending a long time on this and still not getting an correct way.

Regards,
Karthika

[Nickolay V. Shmyrev]

You can start with cmusphinx and continue with kaldi once you have more experience.

[Karthika]

Thanks for you reply Nickolay

I am sorry that,I can't able to understand the meaning of

You can start with cmusphinx and continue with kaldi

Please explain this.Is it possible to use both ??

As I have tried to use the Kaldi also,but is it possible to use that in node js ?

Because still I didn't find that any of the code supporting in node js and Takes more memory also.

The reason I switched to Kaldi is because of the less accuracy in CMUSPhinx.

I don't know whether it is correct to continue in CMUsphinx to improve the accuracy or some other tools like Kaldi ?

As my main aim is to convert the offline/online audio files to text with good Accuracy.

I will be happy,if you give your feedback on this.

Thanks.