Link Search Menu Expand Document Moon Sun
Table of contents
  1. Settings of Rhasspy
    1. 0. Which systems are we using?
    2. 1. MQTT
    3. 2. Audio Recording
    4. 3. Wake Word
    5. 4. Speech to Text
    6. 5. Intent Recognition
    7. 6. Text to Speech
    8. 7. Audio Playing
    9. 8. Dialogue Management
    10. 9. Intent Handling

Settings of Rhasspy

0. Which systems are we using?

We are using the following systems:

We need to set up all of these above, and some more:

I’ll go through the settings with the order on the settings page. You can find the settings page on <ip of your Raspberry Pi>:12101, and then at the gears on the left. To apply all settings you need to click th e save button on the top or bottom, your Raspberry Pi will restart.

1. MQTT

For background information click on the topic header. MQTT is the communication channel rhasspy communicates with. Our MQTT-server is mosquitto we installed in the installation. When you have done all steps, in the installation instruction you have already a running MQTT server on your Raspberry. To use it you choose External in the drop down menu next to the MQTT button in the Rhasspy settings. Now you can set up a hostname, a port, an username and a password. We only need to set up the host and the port. The host ip is the local ip of your Raspberry Pi (127.0.0.1). When you use Hermes Led Control the host ip MUST be an IP and should NOT be the alias localhost!. As port, you need the port 1883. The username and password is only needed if you installed one.

2. Audio Recording

Please make sure you install your microphone correctly! When you installed it correctly, you can proceed. Under the MQTT option is the option Audio Recording. Select arecord in the dropdown menu. In the options below you only need to select the installed microphone (Default Audio Device (sysdefault:CARD=seeed4micvoicec)) or you type sysdefault:CARD=seeed4micvoicec in the upper textbox.

3. Wake Word

We chose Rhasspy Raven as Wake Word system. Select it in the drop down menu. When you open the settings, you first need to record three examples of your Wake Word. You can choose an own one, or you use Sherlock as Wake Word as we did. You see three Record buttons. Press the first one and say your Wake Word, then the second, and then the third. Please make sure you record the Wake Word from 3 different positions in the room for example far away, semi far away and close to the microphone. You can then adjust a few settings.
Probability Threshold is a value between 0 and 1. You can adjust the sensity with it. The lower the value, the more sensitive the rhasspy will react. My best results are with the value 0.52.
Average Templates combines all example wav files to one. This reduces cpu usage, but the accurity will lessen.
Minimum Matches this is a number between 1 and 3. It specifies how many wav files needs to match the detection.

4. Speech to Text

As STT this project is using DeepSpeech 0.7.4, please make sure you have installed it. In the rhasspy config you must set the dropdown menu to Hermes MQTT. Since you installed this custom version of DeepSpeech 0.7.4, you can’t use the option Mozilla DeepSpeech. This is because the default DeepSpeech 0.6.1 of rhasspy can’t load the default pretrained language model and there is no german pretrained language model for this version. We found a good german pretrained language model at AASHISHAG/deepspeech-german, but there is no model for DeepSpeech 0.6.1 so we port rhasspy-asr-deepspeech and rhasspy-asr-deepspeech-hermes to a new version. Transcribing works very good but training does not working. But it’s in development. Please check in your .config/rhasspy/profilesprofile.json if this exists:

"speech_to_text": {
        "system": "hermes"
},

If there are any other line in the curly braces, please delete them.

5. Intent Recognition

For the Intent Recognition you installed Snips-NLU. It is installed as an extra server on your Raspberry Pi. This server starts automatically, so here you just need to choose Hermes MQTT in the rhasspy settings.

6. Text to Speech

Here you can choose your own. We decided to use NanoTTS, it’s simple to use and sounds nice. So choose NanoTTS or one of the TTS you want. When you use nano TTS you only need to set up the language the available languages are en-US, en-GB, de-DE, es-ES, fr-FR and it-IT.

7. Audio Playing

Please choose aplay. When you solved the steps in audio configuration your audio is already configured.

8. Dialogue Management

9. Intent Handling

The Intent Handling is for adding skills to rhasspy. This is realised with Node-RED flows. You can take a look in our skill section.