Voice Goings-On

How Can I Help?

Published

Ever since the Year of Voice started in 2023, the Home Assistant voice community has grown significantly. This growth continues today, with new volunteers reaching out to me almost every day asking how they can help.

While I do my best to tailor responses to each volunteer, I've decided to collect a list of ways to contribute to voice here for reference. I'll make a note when the method requires specific hardware or technical expertise.

Report Problems

Creating GitHub issues helps us know when there's a problem, and helps us track the resolution over time. It can be tricky to know where to file an issue because different parts of the voice stack live in different places. If you're unsure, please ask in one of these places.

For problems with voice commands, the troubleshooting Assist guide can help identify which part of the voice pipeline needs to be fixed. This is important because a microphone issue is very different from a missing voice command.

Help Others

There are always lots of questions related to voice in these places:

If you have the time, please help answer questions or suggest ideas for the community.

Record Wake Word Samples

Our okay nabu wake word is trained from real world recordings, and we need diverse samples! Using your phone or laptop, visit the wake word collective and record yourself saying the wake word as you walk around the room. If members of your family are willing, have them record as well 🙂

Requires: phone or computer microphone

Contribute to Common Voice

Mozilla's Common Voice project is our primary source of data for training speech-to-text models. Speech-to-text is the part of a voice pipeline that transforms your spoken audio into text. Contributing to Common Voice, either by recording yourself speaking or validating other recordings, is an excellent way to contribute to voice projects all around the world (not just Home Assistant).

Requires: phone or computer microphone

Record a Piper Dataset

Piper is a text-to-speech system developed by the Open Home Foundation. This part of the voice pipeline transforms Home Assistant's text response into spoken audio. It runs locally on modest hardware, like the Raspberry Pi 4, and supports a variety of languages.

While many Piper voices were trained from public datasets, some of them were recorded by volunteers using Piper Recording Studio. If you're comfortable contributing your voice to the public domain and you have a few hours to record, we'd be happy to train a new voice!

Requires: a good microphone and a quiet recording area

Become a Language Leader

If you're a native speaker and familiar with GitHub, consider becoming a language leader in order to contribute new voice commands to the intents repository.

Language leaders review and write sentence templates that match Home Assistant's builtin intents to perform actions or answer questions.

Requires: knowledge of version control (git) and template syntax