Rhasspy Raven Wakeword System

Wakeword detector based on the Snips Personal Wake Word Detector.

Dependencies

• Python 3.7
• dtw-python for Dynamic Time Warping calculation
• python-speech-features for MFCC computation
• rhasspy-silence for silence detection

Installation

$ git clone https://github.com/rhasspy/rhasspy-wake-raven.git
$ cd rhasspy-wake-raven
$ ./configure
$ make
$ make install

Running

Record at least 3 WAV files with your wake word. Trim silence off the front and back manually, and export them as 16-bit 16Khz mono WAV files. Then, run:

$ arecord -r 16000 -f S16_LE -c 1 -t raw | \
    bin/rhasspy-wake-raven --distance-threshold D <WAV1> <WAV2> <WAV3> ...

where D is a threshold used to determine if a WAV template matches. A value of 38 for D seemed to work for the sample WAV files in etc/test. You can add --debug to the command line to see the distance values for each match attempt to get an idea for your wakeword.

Example

Using the example files for "okay rhasspy":

$ arecord -r 16000 -f S16_LE -c 1 -t raw | \
    bin/rhasspy-wake-raven --distance-threshold 38 --minimum-matches 2 etc/test/okay-rhasspy-*.wav

This requires at least 2 of the 3 WAV templates to match before output is printed:

{"keyword": "etc/test/okay-rhasspy-00.wav", "detect_seconds": 5.871257066726685, "detect_timestamp": 1594929004.2440836, "raven": {"distance": 35.00761344404474, "threshold": 38.0, "tick": 1, "matches": 3}}

Output

Raven outputs a line of JSON when the wake word is detected. Fields are:

• keyword - path to WAV file template
• detect_seconds - seconds after start of program when detection occurred
• detect_timestamp - timestamp when detection occurred (using time.time())
• raven
  • distance - normalized dynamic time warping distance
  • threshold - threshold used for comparison
  • matches - number of WAV templates that matched
  • tick - monotonic counter incremented for each detection