Customizations

Mycroft is open source and free to customize however you wish. The simplest method for this is by using plugins for common systems.

The Mark II uses the standard Mycroft plugin system for Wake Words, Speech-To-Text (STT), and Text-To-Speech (TTS) - meaning you can install any Mycroft compatible plugin for these services.

Please note, at this time the Mark II does not include a Configuration Manager utility. Any changes must be made directly to your mycroft.conf

STT Plugins

The Mycroft Mark II already includes 3 STT plugins to choose from.

To use a different model, first setup SSH access and edit the Mycroft configuration file at:

/home/mycroft/.config/mycroft/mycroft.conf

After modifying the STT configuration you must restart the Audio Service for changes to take effect:

sudo systemctl restart dinkum-audio

Mycroft STT

This is the default cloud based STT service provided by the Mycroft backend. Your Mark II comes pre-configured to use this option, however if you have changed this, you can revert the configuration by using:

{
  "stt": {
    "module": "mycroft"
  }
}

Currently the "Mycroft STT" service is a proxy for Google Cloud STT as it is the most accurate STT option at this time. Any STT request made to this service is routed through Mycroft's servers to provide additional anonymity. By doing this, we prevent the upstream service from profiling individuals as they cannot see if one device is making 30,000 requests, or if 30,000 devices are making one request each. We have also explicitly disabled data retention by Google for any of these requests.

Despite these protections, we are working to provide an alternative cloud service that does not rely on Google or other services that we do not deem to be privacy-respecting.

We are also working to provide the best on device STT options possible. Currently this consists of two plugins:

Coqui STT

A completely local option utilizing the Coqui STT deep learning toolkit.

{
  "stt": {
    "module": "coqui_stt_plug"
  }
}

Vosk

A completely local option utilizing the Vosk Speech Recognition Toolkit.

{
  "stt": {
    "module": "vosk_stt_plug"
  }
}

TTS Plugins

By default the Mark II comes with the Mimic 3 TTS plugin pre-installed.

The default configuration is currently:

{
  "tts": {
    "module": "mimic3_tts_plug",
    "mimic3_tts_plug": {
      "voice": "en_UK/apope_low",
      "preloaded_cache": "/opt/mycroft/preloaded_cache/Mimic3"
    }
  }
}

The voice model, speaker, speaking rate, variability and other configuration options may be set here. See the Mimic 3 TTS Plugin documentation for more detail:

Mimic 3Mycroft AI

Mimic 3 TTS Plugin configuration options

After modifying the TTS configuration you must restart the Voice Service for changes to take effect:

sudo systemctl restart dinkum-voice

Wake Word Plugins

By default the Mark II comes with the Precise Lite Wake Word Plugin pre-installed.

This plugin runs Precise models that have been converted to TensorFlow Lite, thereby using significantly lower resources than the traditional Precise models.

The same Precise configuration options are available. For more info see:

Using a Custom Wake WordMycroft AI

After modifying any Wake Word configuration you must restart the Voice Service for changes to take effect:

sudo systemctl restart dinkum-voice

Additional Plugins

At this time, we do not have an official list of plugins that have been reviewed by Mycroft. There are a number available on PyPI, however these should be used at your own risk.

To create your own plugins, or find more information on configuring 3rd party plugins - see the main Mycroft documentation for each of the plugin services:

STT PluginsMycroft AI

TTS PluginsMycroft AI

Wake Word PluginsMycroft AI