Kaldi Spotter
wake word spotting with kaldi
Usage
Config
Configuration is loaded from the following locations, with each location overriding values from the previous one
"/opt/kaldi_spotter/kaldi_spotter.conf"
"~/.kaldi_spotter/kaldi_spotter.conf"
Sample Config
{
"listener": {
"default_volume": 150,
"default_aggressiveness": 2,
"default_model_dir": '/opt/kaldi/model/kaldi-generic-en-tdnn_250',
"default_acoustic_scale": 1.0,
"default_beam": 7.0,
"default_frame_subsampling_factor": 3,
},
"hotwords": {
# full sentences
# CRITERIA
# - fairly accurate,
# - important enough to want offline functionality
# - worth answering even if speech not directed at device
"lights on": {
"transcriptions": ["lights on"],
"intent": "turn on lights",
"rule": "equal"
},
"lights off": {
"transcriptions": ["lights off"],
"intent": "turn off lights",
"rule": "equal"
},
"time": {
"transcriptions": ["what time is it"],
"intent": "current time",
"active": false,
"rule": "equal"
},
# wake words
# PROTIP: just run the live demo and see which transcriptions come up
"hey computer": {
"transcriptions": ["hey computer", "a computer", "they computer"],
"sound": "/home/pi/start_listening.wav",
"intent": "listen"
}
}
}Python
see examples
from kaldi_spotter import KaldiWWSpotter
def print_hotword(event):
print("HOTWORD:", event)
def print_utterance(event):
print("LIVE TRANSCRIPTION:", event)
listener = KaldiWWSpotter()
listener.on("transcription", print_utterance)
listener.on("hotword", print_hotword)
listener.run()CLI
Start kaldi spotter from cli
python -m kaldi_spotterSample Output
pi@raspberrypi:~ $ python -m kaldi_spotter
INFO:root:Loading model from /opt/kaldi/model/kaldi-generic-en-tdnn_250 ...
INFO:root:audio source: seeed-4mic-voicecard Multichannel
INFO:root:Listening
('LIVE TRANSCRIPTION:', '{"data": {"confidence": 0.7698909431952632, "utterance": "hey computer"}, "type": "transcription"}')
('HOTWORD:', '{"data": {"hotword": "hey computer", "utterance": "hey computer", "intent": "listen"}, "type": "hotword"}')
('LIVE TRANSCRIPTION:', '{"data": {"confidence": 0.7663563699360755, "utterance": "what time is it"}, "type": "transcription"}')
('HOTWORD:', '{"data": {"hotword": "time", "utterance": "what time is it", "intent": "current time"}, "type": "hotword"}')
Install
System Requirements
sudo apt-get install libatlas-dev pulseaudio-utils pulseaudio cythonPre-trained models
You can install English model kaldi-chain-zamia-speech-en or german model kaldi-chain-zamia-speech-de
Refered as kaldi-chain-zamia-speech-XX bellow
Raspbian 9 (stretch) on a Raspberry Pi 2/3
echo "deb http://goofy.zamia.org/repo-ai/raspbian/stretch/armhf/ ./" >/etc/apt/sources.list.d/zamia-ai.list
wget -qO - http://goofy.zamia.org/repo-ai/raspbian/stretch/armhf/bofh.asc | sudo apt-key add -
apt-get update
apt-get install kaldi-chain-zamia-speech-XXDebian 9 (stretch, amd64)
apt-get install apt-transport-https
echo "deb http://goofy.zamia.org/repo-ai/debian/stretch/amd64/ ./" >/etc/apt/sources.list.d/zamia-ai.list
wget -qO - http://goofy.zamia.org/repo-ai/debian/stretch/amd64/bofh.asc | sudo apt-key add -
apt-get update
apt-get install kaldi-chain-zamia-speech-XXCentOS 7 (amd64)
cd /etc/yum.repos.d
wget http://goofy.zamia.org/zamia-speech/misc/zamia-ai-centos.repo
yum install kaldi-chain-zamia-speech-XXPip Package
install from pip
pip install kaldi_spotteror from source
pip install git+https://github.com/JarbasAl/kaldi_spotter
