Open source speech recognition

2009 August 30
tags: , , , ,
by Liz Henry
Open source speech recognition

Open source speech recognition software needs large samples of transcribed speech recording to make up an acoustic model. The VoxForge project is building acoustic models for English, German, Spanish, French, Hebrew, Italian, Dutch, Portuguese, and Russian. You can contribute to one of these languages from your computer, recording and uploading to VoxForge’s site, or by telephone. It looks very easy to contribute, and it will then be licensed under the GPL.

Most acoustic models used by ‘Open Source’ speech recognition (or Speech-to-Text) engines are ‘Closed Source’. They do not give you access to the speech audio and transcriptions (i.e. the speech corpus) used to create the acoustic model.

The reason for this is that Free and Open Source (‘FOSS’) projects are required to purchase large speech corpora with restrictive licensing. Although there are a few instances of small FOSS speech corpora that could be used to create acoustic models, the vast majority of corpora (especially large corpora best suited to building good acoustic models) must be purchased under restrictive licenses.

from → open source, software

6 Responses
  1. Luis permalink
    August 30, 2009

    Hi,

    What is the speech recognizer used in this video?
    Is it possible to send me a link with informations regarding this system (software, scripts used to train acoustic models, etc)?

    I would like to try build some models for Brazilian Portuguese.

    Thanks,

    Luis

  2. Liz Henry permalink
    August 30, 2009

    Hi Luis, the speech recognizer in this video is GnomeVoiceControl, http://live.gnome.org/GnomeVoiceControl and the person who posted the video has some more information here, in Spanish, http://www.elrinconde.com/mimetist/2008/12/23/gnome-voice-control-controla-tu-escritorio-con-la-voz/

    And there are also these projects mentioned on VoxForge:
    Sphinx, ISIP, Julius e HTK (nota: HTK possui restrições de distribuição).

    The VoxForge models for Brazilian Portugues are being built here,

    http://voxforge.org/pt_br

    and there is a forum, but without much activity yet. You might try posting there:

    http://voxforge.org/pt_br/forums

    I’ve heard of the Center for Independent Living in Brazil, CVRJ,

    http://www.independentliving.org/docs3/cils.html#anchorBra

    But their site appears to be down.

    Good luck and let us know about progress on your project!

  3. Nic permalink
    September 2, 2009

    This looks promising. I’d been looking for a voice recognition solution to run on Ubuntu for a while, but nothing much was available. Ideally, I’d love a FOSS solution I could run both on *nix and Windoze, using the same user files/dictionaries. I don’t want much at all ;)

  4. Liz Henry permalink
    September 2, 2009

    Hey Nic,

    I came across this too,

    http://dot.kde.org/2009/08/22/simon-speech-activated-user-interface-kde

    Not sure how far along the project is, but it looks neat.

    Are you going to linux.conf.au? I’m going to be there and am speaking on DIY assistive tech.

  5. Nic permalink
    September 2, 2009

    Thanks for the link to simon. I shall investigate :)

    I’m waiting to hear back if my paper will be accepted for linux.conf.au. I hope so, since I was invited to send my paper in by one of the organisers. But one never knows :)

  6. Amy permalink
    June 3, 2010

    Thanks for the link to simon. I shall investigate :)

    I’m waiting to hear back if my paper will be accepted for linux.conf.au. I hope so, since I was invited to send my paper in by one of the organisers. But one never knows :)

Comments are closed.