Skip to content

Conversation

@Navbryce
Copy link

  • Implemented a hotword detector (Listens for "Klauba") before recording and parsing speech input
  • "Intelligent" voice recording after "Klauba" is detected
    • Wont' start recording until silence has been broken
    • Stops recording after a certain amount of continous silence
  • Improved accuracy of speech recognition
    • Fine-tuned the corpus (and regenerated language and lexical models)
    • Created a script to autogenerate corpuses based off of sententence templates and room numbers
    • Removed the pre-input noise filtering code (in the previous iteration, we noise filtered the code BEFORE we passed it to shpinx)
  • Parsed speech now passed to a python script, which can distribute the text to all other code (like the movement code)
  • Script won't stop running until stopped by the user

Navbryce and others added 26 commits October 20, 2019 19:12
…updated decode.c to use default audio device, how the record rate was specified, and made the paths more generic (shoul be replaced by an environment variable in the future
… with snowboy. interestingly, the snowboy terminate method hangs when called
…ntinous silence. Now waits for a continous amount of silence. fixed a bug that inputted the raw audio instead of the filtered audio. decode script now infinitely loops until the user hits cntrl-c
…ller duration of silence before recording stops). Explicitly enabled noise reduction when calling sphinx. Used the raw recording instead of the noise filtered version to pass to sphinx.
@jiashenC
Copy link
Member

Can you clean the commits little bit? Seems like so many files.

@Navbryce
Copy link
Author

Navbryce commented Dec 12, 2019

I don't want to .gitignore or remove any of the files in the pocketsphinx or cmusphinx directories because I'm not sure which ones are required for the libraries to build.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants