Menu

Speech Recognition on Portuguese (Portugal)

Help
2017-05-19
2017-05-19
  • Hugo Alexandre

    Hugo Alexandre - 2017-05-19

    We (Me and a colleague) are developing speech recognition on our native language. We followed all the tutorial, and finished a few days ago the building of our accoustic model. We have 15 words, and collected 4 recordings from different people. The results are not accurate, the model only points out 2 correct words in 15 words total.

    I'm posting this in order to understand, if possible, what can we do to improve the model acccuracy.

    Grateful for your atention,

    Hugo Alexandre

     
    • Nickolay V. Shmyrev

      Hello Hugo

      This is a good start.

      Our tutorial says you need much more data to train a good model, you can collect lectures, maybe audiobooks, not necessary transcribed, you can also check radio shows or podcasts. You need 300+ hours of data from 50-100 speakers.

      Once you collect the data you can contact me (nshmyrev@gmail.com), I'll help you to train a good model.

       

Log in to post a comment.

Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.