Link Search Menu Expand Document
Table of contents
  1. Rhasspy configuration
    1. Rhasspy technology stack
      1. Rasa natural-language understanding *
    2. Auto configuration
    3. Manual configuration
      1. Audio recording
      2. Audio playing
      3. Wake Word
      4. Speech to Text
      5. Intent Recognition
      6. Text to Speech
      7. Dialogue Management

Rhasspy configuration

Rhasspy technology stack

Find below the technology stack used for the upcoming Rhasspy configuration. One of the goals for this project was to evaluate usability of said technology stack. Evaluation results can be found here.

  • Wake Word: Pocketsphinx
  • Speech to Text: Pocketsphinx
  • Intent Recognition: Rasa NLU * - Fsticuffs
  • Text to Speech: NanoTTS
  • Dialogue Management: Rhasspy

Rasa natural-language understanding *

Rasa NLU was intended to be used in this project. Switching to Fsticuffs was necessary because of various documented reasons. Fsticuffs uses the rhasspy-nlu library and does not require additional configuration. If you are interested in Rasa NLU, check out our Rasa NLU page here.

Auto configuration

Desired configuration can be achieved more quickly by simply copying the profile settings into Rhasspy’s configuration file.

cd /home/pi/rhasspy/.config/rhasspy/profiles/de
nano profile.json

{
    "dialogue": {
        "system": "rhasspy"
    },
    "intent": {
        "system": "fsticuffs"
    },
    "microphone": {
        "pyaudio": {
            "device": "1"
        },
        "system": "pyaudio"
    },
    "sounds": {
        "aplay": {
            "device": "plughw:CARD=Headphones,DEV=0"
        },
        "system": "aplay"
    },
    "speech_to_text": {
        "system": "pocketsphinx"
    },
    "text_to_speech": {
        "system": "nanotts"
    },
    "wake": {
        "pocketsphinx": {
            "keyphrase": "hey apollo"
        },
        "system": "pocketsphinx"
    }
}

Press CTRL + x followed by y and then hit enter to save all changes.

Restart Rhasspy service.

cd /home/pi/rhasspy
docker-compose down
docker-compose up -d

After the restart access Rhasspy’s web interface and follow instruction for downloading required Pocketsphinx profile files and confirm pronunciation for unknown words used in Rhasspy’s example intents.

Manual configuration

Access Rhasspy’s web interface settings page at ip-address:12101/settings.

Audio recording

  1. Select PyAudio
  2. Press Save settings
  3. Click on Audio Recording and press refresh. Rhasspy is going to select your attached ReSpeaker 4-Mic Array. Otherwise a manual selection is required.
  4. Press Save settings

Audio playing

  1. Select aplay
  2. Press Save Settings
  3. Click on Audio Playing and then choose Hardware device with all software conversions from Available Devices.
  4. Press Save Settings

Wake Word

  1. Select Pocketsphinx
  2. Press Save Settings
  3. Click on Wake Word and assign a desired wake word.
  4. Press Save Settings

Speech to Text

  1. Select Pocketsphinx
  2. Press Save Settings
  3. Click on Download to download required Pocketsphinx profile files
  4. Click View and then Confirm Guesses to accept pronunciation for unknown words used in Rhasspy’s example intents.
  5. Press Restart in the top right corner

Intent Recognition

  1. Select Fsticuffs
  2. Press Save Settings

Text to Speech

  1. Select NanoTTS
  2. Press Save Settings

Dialogue Management

  1. Select Rhasspy
  2. Press Save Settings

Continue with skill implementation