0

I'm using wit.ai as a (probably) temporary solution for a speech-to-text service in my own personal assistant. I then pass along this transcribed text elsewhere for further analysis. Eventually I'll probably move away to something a little more hefty (Watson, PocketSphinx, etc.), but wit.ai seems a good first step because it's free and easy.

However, I'd like to add my own custom words to wit.ai. Specifically names of bands that aren't real English for my music player.

Ideally I'd be able to upload a large file or hit an api many times with example phrases that it could expect.

Something like:

Play Black Dog by Led Zeppelin
Play Come Down by Anderson Paak
Play Hippie Babysitter by Papadosio

Words like Paak and Papadosio aren't in the dictionary but can be sounded out just like any word. I know CMUSphinx/PocketSphinx can do this just fine. Does anyone know of a way that Wit.ai can do this?

benwiz
  • 2,167
  • 3
  • 22
  • 33

1 Answers1

0

I don't think it's possible to add your own audio data in wit.ai (not sure about this but I'd bet you can't, being a closed-source propietary system). EDIT: You may correct your audio transcriptions in the Inbox > Audio section of their online interface, that might be enough for your needs.

You also mentioned CMU Sphinx; they accept audio contributions in order to improve the available languages, through VoxForge.

Another option would be the Google Speech Recognition API, which has a really extensive audio corpus, providing really great results. You may want to try it out (you can test it online in the link) with your custom band names and see if it fits your needs.

scb
  • 46
  • 4