mirror of
https://github.com/OpenSquawk/OpenSquawk
synced 2026-08-02 06:46:18 +08:00
Until now the Speaches branch sent every request to the single env-configured model (de_DE-thorsten — a German voice for English radio), so the voice pool had no audible effect. A server-side registry now resolves the logical voice ids to disjoint controller/pilot Piper speakers; unknown ids keep the env fallback and the cache key follows the resolved pair. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>