About speech recognition online

[morneau]

I'm a newbie to Sphinx. We intend to develop an English learning website, not
sure if we can use Sphinx to implement it. It will be a speaker independent
application for students to learn and practice spoken English online, with
following functions:

1. word and sentence speaking assessment: The application displays a word or sentence for the speaker to speak and record, then the application will assess the speaker's pronunciation accuracy.

2. dialog recognition: The application already defined some dialog between two people. The speaker and the application simulate the dialog by talking to each other. Such the application should recognise which sentence that the speaker just spoke (the speaker must speak one of the pre-defined sentences), then speaks next sentence to the speaker accordingly.

So, has Sphinx the capability to implement the above? What will be the
recognition accuracy? Is there any existing application similar to our
requirement? Or is there any document to teach me how to start our task
quickly?

Any help would be appreciated.

Morneau

[Nickolay V. Shmyrev]

Hello

So, has Sphinx the capability to implement the above?

CMUSphinx engine is a good basement for the implementation above. Though such
features are not readily implemented yet.

What will be the recognition accuracy?

It more depends on size and the quality of the data you will collect. Then it
depends on the implementation too. You can get a good working system if you
put enough resources into it.

Is there any existing application similar to our requirement?

There are services built with CMUSphinx with similar functionality, but no
open source applicaiton I know

Or is there any document to teach me how to start our task quickly?

Most of the issues covering the tasks above (utterance verification and
pronunciation assessment) are described in FAQ:

http://cmusphinx.sourceforge.net/wiki/faq#qcan_pocketsphinx_reject_out-of-
grammar_words_and_noises
http://cmusphinx.sourceforge.net/wiki/faq#qhow_to_implement_pronunciation_eva
luation

[morneau]

Hi nshmyrev,

1. "There are services built with CMUSphinx with similar functionality, but no open source applicaiton I know": Do you have the URL of such online service?

2. I installed Sphinx-4 on Windows 7 and tried the Hello World demo. But the recognition accuracy is pretty bad. I think the reason is this demo uses WSJ Acoustic Model but I am not native English speaker. If change to use voxforge Acoustic Model, the recognition accuracy would improve. Am I correct?

[Nickolay V. Shmyrev]

Do you have the URL of such online service?

Yes

I think the reason is this demo uses WSJ Acoustic Model but I am not native
English speaker.

There could be other reasons too

If change to use voxforge Acoustic Model, the recognition accuracy would
improve. Am I correct?

Unlikely

[sahlathasnim]

we are going to do a project on auto subtitles to movies.for that we need to
have a cmu sphinx package to translate speech to text.can anyone suggest a
link to download that.

[Nickolay V. Shmyrev]

for that we need to have a cmu sphinx package to translate speech to
text.can anyone suggest a link to download that.

you can download CMUSphinx here

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

[morneau]

nshmyrev, it would be appreciated if you can share some URL of existing online
services using CMU Sphinx.

[Nickolay V. Shmyrev]

if you can share some URL of existing online services using CMU Sphinx.

http://nexiwave.com