Spaces:
Sleeping
Sleeping
Commit History
-
Revert "feat: optional server-side background ambience mix"
This reverts commit a13f34e4846a3951f661530745fb88b6d0a639c2.eabcac1
-
feat: optional server-side background ambience mix
Add a `background` + `background_snr` field to TTSRequest (and the /tts/clone form) so any endpoint can return speech pre-mixed with a looped ambience bed at a target SNR. Backend stays clean by default (background=null). Bundles four 30s mono-24kHz beds (call-center, coffee-shop, office, retail), tracked via Git LFS, and ships them into the Docker image. Mixing is a pure numpy post-step on the int16 PCM. Note: per-utterance/stateless — bed restarts each call; a seamless live call still wants a continuous loop in the client audio layer. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>a13f34e
-
Pin torch==2.8.0 + transformers==5.3 to fix higgs_audio_v2_tokenizer import
Verified signature6373f05 -
Self-hosted OmniVoice TTS on T4 (FastAPI, no proxy)
Verified signaturedcf73a3
