-
-
Notifications
You must be signed in to change notification settings - Fork 47
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Speaker Recognition #367
base: naomi-dev
Are you sure you want to change the base?
Speaker Recognition #367
Commits on Sep 15, 2022
-
Working on introducing a new speaker recognition plugin type. This plugin is to both recognize an individual and taylor responses to them and improve speech recognition by training STT models to individual speakers.
Configuration menu - View commit details
-
Copy full SHA for 421d062 - Browse repository at this point
Copy the full SHA 421d062View commit details
Commits on Oct 17, 2022
-
Added a default speaker recognizer
Added a default speaker recognizer, default_sr, which does not attempt to identify the speaker but just responds with the first name as stored in the profile. I am also passing the result around from the sr_plugin to the intent parser, to which the identity of the speaker can be attached to the intent object being passed to the speechhandler. I also simplified the list of parameters being passed to the mic object when it is created, storing them instead in the profile.
Configuration menu - View commit details
-
Copy full SHA for 26a663a - Browse repository at this point
Copy the full SHA 26a663aView commit details
Commits on Nov 6, 2022
-
Fixed problems with speech recognizer
There were numerous problems with the speech recognizer class that made it not work when using the VOSK_sr plugin. I am no longer trying to recognize the speaker when listening passively for the wake word, only when doing the active listening. This is because the passive listening needs to be very fast. I am now putting the name of the identified user in parenthesis after the active listening transcript. Plugins can access the identity of the speaker as `intent.get('user', '')`. The only plugin currently set up to use this is the shutdown plugin. I also have an update to the Greetings plugin which greets you by name when you greet Naomi. The setup still assumes en-US when downloading the VOSK models, which needs to be fixed to respect the "language" setting in the profile. The VOSK speaker recognition is not terribly accurate. It also seems like you need to retrain your speaker recognition database from new recordings when you switch to different recording hardware.
Configuration menu - View commit details
-
Copy full SHA for dab0b66 - Browse repository at this point
Copy the full SHA dab0b66View commit details
Commits on Nov 7, 2022
-
Fixed some method signature mismatches
Fixed some method signature mismatches from when I removed the parameters from the mic methods. Fixed an issue preventing the input device verification from working during initial setup or repopulate. Changed the name of the "confidence" result from speaker recognition to "distance" (since smaller numbers are better). Clarified how to enter multiple email addresses in notification client configuration, although I think that still needs looked at. I think the safe email list is not being stored as a list, and I think Naomi should only respond to email addresses in that list, not to any email address if the list is empty.
Configuration menu - View commit details
-
Copy full SHA for 1af470c - Browse repository at this point
Copy the full SHA 1af470cView commit details -
Download French or German language models
Added filenames for the default French and German models. Constructed URLs and paths from these models to automatically download and extract the model that matches the language choice in profile. The only thing it does not currently do is check whether or not VOSK is also used as the STT engine. If not, then the audio file pointer should be passed to the actual STT engine.
Configuration menu - View commit details
-
Copy full SHA for fbc8f46 - Browse repository at this point
Copy the full SHA fbc8f46View commit details
Commits on Nov 8, 2022
-
Fix for when mic.listen returns False
If the mic gets cut off when Naomi is listening, then the mic.listen() method will return False rather than an sr_output dictionary.
Configuration menu - View commit details
-
Copy full SHA for f8c1106 - Browse repository at this point
Copy the full SHA f8c1106View commit details