Problem statement
Speech recognition is one of the main reported problems consistently - and solving it has been hard to prove. One main issue (but not the only one) is that auto gain can do as good as harm, depending on the situation. But there is an opportunity to improve this with a better Speak to text (STT).
For expectations, there is an important chance this makes the Voice experience better, but its not a single solution - just another brick in the wall.
Community signals
Community survey, although biased towards high-tech users and not clear on a specific problem, showed how there is not a great satisfaction towards voice recognition + achievement of the task. Its not conclusive, but grants the chance to at least explore this opportunity
Scope & Boundaries
In scope
- New STT
- Being able to test it with and without auto gain
Not in scope
- Bigger architectural changes to Voice
Foreseen solution
Add second audio channel for voice
esphome PR - esphome/esphome#16265
aioesphomeapi PR - esphome/aioesphomeapi#1625
Risks & open questions
- Is the solution really better?
- Can we test it propperly?
- How many biases do we meet when testing?
Appetite
Small - Should be done in one cycle of 2 releases.
Execution issues
No response
Decision log
Problem statement
Speech recognition is one of the main reported problems consistently - and solving it has been hard to prove. One main issue (but not the only one) is that auto gain can do as good as harm, depending on the situation. But there is an opportunity to improve this with a better Speak to text (STT).
For expectations, there is an important chance this makes the Voice experience better, but its not a single solution - just another brick in the wall.
Community signals
Community survey, although biased towards high-tech users and not clear on a specific problem, showed how there is not a great satisfaction towards voice recognition + achievement of the task. Its not conclusive, but grants the chance to at least explore this opportunity
Scope & Boundaries
In scope
Not in scope
Foreseen solution
Add second audio channel for voice
esphome PR - esphome/esphome#16265
aioesphomeapi PR - esphome/aioesphomeapi#1625
Risks & open questions
Appetite
Small - Should be done in one cycle of 2 releases.
Execution issues
No response
Decision log