Docker-based web interface for generating .pmdl files from audio examples for snowboy.
Credit to seasalt-ai
To get started, run the pre-built Docker image:
$ docker run -it -p 8000:8000 rhasspy/snowboy-seasaltThe web interface will now be available at http://localhost:8000
If you'd like to build the Docker image yourself, see scripts/build-docker.sh
Examples can be recorded and submitted directly in a web browser. Once you've enabled your microphone (required by Chrome), each example can be recorded by:
- Clicking the "Record" button and waiting until it's ready
- Speaking the wake word and then quietly waiting until it's done
The timeouts here are used to avoid needing to trim out button clicks from the audio.
Once you have 3 examples, click "Submit" and you should see a "Save Model" button appear. Save the .pmdl file somewhere and enjoy your new wake word!
Audio visualization is provided by wavesurfer-js
You can also POST your audio examples to the web server directly and get a .pmdl file back!
The /generate endpoint expects form data with:
- A
modelNamefield with your wake word name - At least 3 files with audio samples
- These will be converted to the appropriate format with
ffmpeg - Unless
?noTrim=trueis given, silence is automatically trimmed from the beginning and end of each example
- These will be converted to the appropriate format with
$ curl \
-X POST \
-F modelName=okay-rhasspy \
-F [email protected] \
-F [email protected] \
-F [email protected] \
--output okay-rhasspy.pmdl \
localhost:8000/generate