Official documentation
Next iteration of our Voice Assistant is here (Voice chapter 10)
Home Assistant project, Home Assistant blog, 2025
Read the originalRead on October 17, 2026
- What it is
- The June 2025 Home Assistant blog post on the voice project's progress, including why Speech-to-Phrase cannot pass unexpected commands to a language model and why large Whisper models need a GPU for some languages.
- How much weight it carries
- The project's own account of its design trade-offs, written by the team. It describes the state of the software in June 2025, so later releases may differ.
Facts taken from this source
- Home Assistant says you cannot use a language model as a fallback with Speech-to-Phrase, because unexpected commands cannot be converted into text for the model to process.
“because unexpected commands can’t be converted into text for the LLM to process” (Speech-to-Phrase)
- Home Assistant says French may work well enough with the largest Whisper model, which needs a GPU, but is unusable on a Raspberry Pi or even an N100-class PC.
“A language like French may work well enough with the largest Whisper model (which requires a GPU), but is unusable on a Raspberry Pi or even an N100-class PC.” (Language support)