A pre-release for field validation. HACS offers this only to users who have enabled beta versions for this repository. It is not a public release, and the changes below still ship to everyone in the next monthly bundle as 3.43.0.
It supersedes v3.43.0-beta.7 for #671 testing. Please use this one rather than beta.7.
What changed since beta.7
Multi-sentence speech from a local server is no longer cut off (#689). Speaches with a piper voice answers a request of several sentences with one WAV file per sentence, back to back, and each file's header gives only its own sentence's length. beta.7 read only the first file:
- Streamed replies: every chunk of a reply lost the sentences after its first. When that first sentence was "Yes." or "Sure!", you heard a fraction of a second of speech.
- Announcements: a player that asks for WAV heard only the first sentence. This has been the case since 3.38.0.
Every sentence is now read, and a whole message is merged into one file. With real Speaches clips, 0.92 s → 2.79 s and 0.63 s → 2.95 s of audio are now recovered.
A note on the acknowledgement phrase
Some voices release a final hard consonant as a separate little burst after a short pause. piper's "moment" ends "…momen", about 25 ms of silence, then "t". Inside a sentence the next word covers it. As an acknowledgement, a pause follows while the model works, so the detached "t" (or "k" in "Let me check.") can sound like speech that started and was cut off. A phrase ending in a softer sound avoids it, for example One moment, please. or Let me see.
Everything from beta.7 and earlier is included. See CHANGELOG.md under Unreleased.