0

I am using Sphinx4 to convert speech to text, it works fine for English, but how can I use it for Hindi speech. For example i said "Mera name amit hai". After converting to text the output should be "मेरा नाम अमित है"

Nikolay Shmyrev
  • 24,897
  • 5
  • 43
  • 87
Amit
  • 55
  • 2
  • 8
  • have you got the speech to text converter for Hindi? I have the same requirement can you guide, How did you implement that? – Devendra Apr 04 '17 at 09:14

1 Answers1

1

You can download Hindi acoustic model to use it with sphinx4 here:

https://sourceforge.net/projects/hindiasr/files/Hindiasr/HindiASR-2.0/

However, it's in a very basic state and it's unlikely it will have high accuracy. For better accuracy you might want to try an acoustic model yourself, see the tutorial for details

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

The setup of the model in Eclipse is simple:

1) download the latest code as described in the tutorial:

http://cmusphinx.sourceforge.net/wiki/tutorialsphinx4

2) Open transcriber demo and edit path to the acoustic model and the dictionary to point to models downloaded from hindiasr

3) Create a grammar or a language model as described in lm tutorial

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

4) Replace the file for recognition in TranscriberDemo.java.

5) Run and see it recognizes the file

Nikolay Shmyrev
  • 24,897
  • 5
  • 43
  • 87
  • thankyou nikolay shmyrev....I already searched about HindiASR-2.0 but i don't know how to configure it? i tried but failed. Is there any video about it ? or if possible please tell me step by step process to run and configure this HindiASR-2.0 in Myecllipse. – Amit May 07 '14 at 05:30
  • hi.. I tried but there is a problem in the import edu.cmu.sphinx.api.Configuration; import edu.cmu.sphinx.api.SpeechResult; import edu.cmu.sphinx.api.StreamSpeechRecognizer; import edu.cmu.sphinx.result.WordResult; the sphinx does not have this api. there is folder of sphinx which have all the API's java file but not the class file I also tried to compile the java file to get the class file . but it shows error in all file except Configuration.java file. – Amit May 09 '14 at 07:02
  • It should work out of box, you checkout the sources, import them as a project into Eclipse and compile them. Try again and don't do anything yourself. Provide more information on what you have done and what issue do you face. – Nikolay Shmyrev May 09 '14 at 08:01
  • I imported as a project into Eclipse and the Transcriber working fine. But i don't know which files or folders in HindiASR are for acoustic model and the dictionary. – Amit May 09 '14 at 09:38
  • Model folder is hindi/model_parameters/hindi.cd_cont_1000. Dictionary is hindi/etc/hindi.dic – Nikolay Shmyrev May 09 '14 at 23:31
  • It gives an error Exception in thread "main" Property exception component:'acousticModelLoader' property:'location' - Can't locate resource:/models/acoustic/hindi.cd_count_1000 – Amit May 10 '14 at 11:21
  • and Exception in thread "main" Property exception component:'null' property:'null' - java.io.FileNotFoundException: models\acoustic\hindi.cd_count_1000\means (The system cannot find the path specified) – Amit May 10 '14 at 11:29
  • You need to specify the path properly, it can be either a full path or a path to the resource within class path, relative path can also work but you need to make sure that relative path is specified agains your current working directory. – Nikolay Shmyrev May 10 '14 at 18:02
  • I set the path properly its working but it gives an error due to feat.params file. There is no feat.params file in HindiASR. HindiASR also have a feat folder but this empty. Exception in thread "main" Property exception component:'null' property:'null' - java.io.FileNotFoundException: models\acoustic\hindi.cd_cont_1000\feat.params (The system cannot find the file specified) Caused by: java.io.FileNotFoundException: models\acoustic\hindi.cd_cont_1000\feat.params (The system cannot find the file specified) – Amit May 13 '14 at 05:55
  • Create empty feat.params in a model folder. – Nikolay Shmyrev May 15 '14 at 05:06
  • Hi Amit, great you got it working! But I want to know, have you tested the accuracy of the acoustic model @NikolayShmyrev suggested? – Akshat Nov 02 '14 at 15:59