brmson/Personal-Assistant

pywit 2.0

PavelTrutman opened this issue · 2 comments

PavelTrutman commented

Today has been released pywit version 2.0, in which they have stopped support for audio recording (see few last comments on wit-ai/pywit#14). Therefore the speech-to-text module is not working with the new version (but it works well with pywit version 1.1).

We have some options:

  1. stay at version 1.1 -> we will be stuck forever
  2. implement audio recording for pywit as described on https://wit.ai/docs/http/20141022#get-intent-via-speech-link
  3. use PocketSphinx for speech-to-text part, use pywit only to get the intent from the sentece

What do you think? Will you look at it?

konrajak commented

I looked into the issue.
I belive that best option would be to use pocketSphinx for speech to text and wit.ai http request only to decode obtained string. Doing this we will avoid sending wav as we did so far. I already asked Petr Kovář to implemet usage of pocketSphinx and I will upload another version of my module.

In the meantime I recommend staying on pywit 1.1. That should also be our backup option on case of problems as it worked well so far and I don't see much benefit in upgrading to 2.0

PavelTrutman commented

Yesterday has been released pywit version 3.0, which is not backward compatible with the version 2.0. Please, look at it.