Speech Recognition Engines

Flat
Julius 4.3.1 decoder high WER
User: abdosalem
Date: 1/3/2016 5:32 am
Views: 6484
Rating: 0

Dears,

I installed Julius 4.3.1 on windows/Cygwin and used my own arabic acoustic model HTK 3.4, corpus 8.5 hours, ngram LM, tied.list. The decoder is working well, however, the results is far away from HTK HDecode. I have HDecode HTK WER = 30% while in julius it is 90% which i think there is something wrong in julius decoder configuration.

The configurations are:

-input mfcfile

-filelist filelist/test.scp

-nlr test.lm

-v test.dic

-h hmmsdef.mmf            # acoustic HMM (ascii or Julius binary)
-hlist tied.list        # HMMList to map logical phone to physical

-b 10000

-s 20000             # hypotheses stack size on 2nd pass (#hypo)
-m 10000            # hypotheses overflow threshold (#hypo)
-n 100                # num of sentences to find

I also applied the LM scale factore (10) that i'm using in HTK as follows:

-imp 10.0 0.0

If you kindly help tweaking the decoder as i think it should give closed WER to HTK.

Appreciate your response.

Abdo

--- (Edited on 1/3/2016 5:32 am [GMT-0600] by abdosalem) ---

Re: Julius 4.3.1 decoder high WER
User: TonyR
Date: 1/3/2016 6:06 am
Views: 85
Rating: 0

Julius has always made search errors and has never achieved the accuracy of HDecode.  However, as you say, the difference shouldn't be that great.  Julius has moved to github and is supported again - https://github.com/julius-speech/julius - I suggest you ask there.

 

Tony

--
Speechmatics is a trading name of Cantab Research Limited
We are hiring: www.speechmatics.com/careers
Dr A J Robinson, Founder, Cantab Research Ltd
Phone direct: 01223 794096, office: 01223 794497
Company reg no GB 05697423, VAT reg no 925606030
51 Canterbury Street, Cambridge, CB4 3QG, UK

 

--- (Edited on 3-January-2016 12:06 pm [GMT+0000] by TonyR) ---

Re: Julius 4.3.1 decoder high WER
User: Abdo
Date: 1/3/2016 8:51 am
Views: 3015
Rating: 0

Thank you for your high response and advice

--- (Edited on 1/3/2016 8:51 am [GMT-0600] by Visitor) ---

PreviousNext