Skip to content

Latest commit

 

History

History
56 lines (42 loc) · 1.33 KB

voice-assistant.md

File metadata and controls

56 lines (42 loc) · 1.33 KB
Raw

Voice Assistant

In order to have the experience of Voice Assistant, the following approach is done:

flowchart TD
    S(Speaker) --> P
    P[Porcupine] --> |wake word| R(Record voice)
    R --> H(Home Assistant)
    H -->|STT| O(OpenAI Whisper)
    O --> |Transcribed text| S
    S --> |Intent| H

Setup

Create and configure the following files in the speaker:

/data/listener

HA_URL="https://hass.example.net"
HA_TOKEN="ey...."
# change if using any other
HA_STT_PROVIDER=stt.faster_whisper
STT_RATE=16000

# use any of Porcupine available
# https://github.com/Picovoice/porcupine/tree/v1.9/resources/keyword_files/raspberry-pi
WORD=alexa
STT_LANGUAGE=en

/data/tts.conf

LANGUAGE=en

You will hear back the transcribed message with TTS.

Troubleshooting

If you always receive error messages, check that date is correct, as it can cause TLS connection issues. You can sync time manually:

ntpd -q -p pool.ntp.org

Find any transcribed content at /tmp/stt.log , along with /tmp/stt.json status.

If you want to change the language for Speech-to-Text:

  • Change STT_LANGUAGE config in the speaker
  • Run /etc/init.d/listener restart or mute and un-mute the speaker
  • Change the language in Home Assistant STT Whisper Addon
  • Restart Home Assistant to refresh the new STT entity settings