This repository contains training data for Home Assistant's local voice control. For a list of supported languages and it's progress, see the website.
Repository layout:
sentences/<language>
- YAML files for
<language>
with the name<domain>_<intent>.yaml
- File format
- YAML files for
responses/<language>
- YAML files for
<language>
with responses for intents - File format
- YAML files for
tests/<language>
- YAML files for
<language>
with test sentences and corresponding intents - File format
- YAML files for
See the documentation for more information.
Each language is maintained by one or more language leaders. Language leaders are responsible for reviewing the contributions in their language and making sure that they are grammatically correct.
Anyone can apply to become one. If you want to apply to be a language leader, join us in #devs_voice
on Discord or open an issue.
Anyone can contribute to the repository. The sentences will be reviewed by the language leaders and merged if they are correct. You can either contribute new sentences or improve existing ones.
If you only want to contribute sentences that should be supported, but don't feel comfortable with YAML, you can add them to the issue of your language. Limit your submissions to commands that work with the supported intents.
See intents.yaml for the Home Assistant intent schemas that are supported.
Home Assistant will automatically generate the following lists during recognition:
{name}
- List of device or entity names
{area}
- List of area names
Checkout the repository and get a development environment with script/setup
. This will create a new virtual environment in the venv
directory of the repository, and install all necessary requirements.
Before developing, always activate your virtual environment with source venv/bin/activate
.
Validate that the data is correctly formatted:
python3 -m script.intentfest validate --language nl
Run the tests. This will parse the sentences and verifies them with the test sentences.
pytest tests --language nl -k fan_HassTurnOn
Leave off --language
to test all languages. Leave off -k
to test all files.
You can try parsing sentences for a specific language with:
python3 -m script.intentfest parse --language en --sentence 'turn on the lights in the kitchen'
This will print a line of JSON for each --sentence
:
{
"text": "turn on the lights in the kitchen",
"match": true,
"intent": "HassTurnOn",
"slots": {
"area": "kitchen",
"domain": "light"
}
}
You can sample the possible sentences for a specific language with:
python3 -m script.intentfest sample --language en -n 1
This will print a line of JSON for each possible sentence:
python3 -m script.intentfest sample --language en -n 1
{"intent": "HassTurnOff", "text": "turn off all the fan in the kitchen"}
{"intent": "HassTurnOn", "text": "turn on the light in the kitchen"}
{"intent": "HassCloseCover", "text": "close the bedroom lamp"}
{"intent": "HassClimateSetTemperature", "text": "set the temp to 0 degrees celsius"}
{"intent": "HassLightSet", "text": "set the bedroom lamp brightness to 0 percent"}
{"intent": "HassOpenCover", "text": "open the bedroom lamp"}
{"intent": "HassClimateGetTemperature", "text": "what's the temp "}
You can filter for specific intents by adding --intents HassTurnOn HassTurnOff
.
Leave off -n
to generate all possible sentences.
To quickly test a sentence template, use:
python3 -m script.intentfest sample_template <template>
For example:
python3 -m script.intentfest sample_template 'open [the] door'
open the door
open door
You can add lists, ranges, and expansion rules as well:
python3 -m script.intentfest sample_template 'set color to <color> and brightness to {brightness}' --values color red green --range brightness 1 2 --rule color '[the] {color}'
python3 -m script.intentfest add_language <language code> <language name>
<language code>
should be something like en
or pl
according to ISO 639.
Language name should be the name of the language in its own language.