--- title: ARIA Audio emoji: 🔊 colorFrom: purple colorTo: blue sdk: docker app_port: 7860 pinned: false license: apache-2.0 short_description: MMS-TTS (pol/eng) + voice emotion for ARIA AGENT --- # ARIA Audio Tiny FastAPI service backing the **ARIA AGENT** app for the models that no serverless inference provider hosts: - `POST /tts` `{"text": "...", "lang": "pol" | "eng"}` → `audio/wav` (facebook/mms-tts-pol / facebook/mms-tts-eng) - `POST /emotion` (body: 16 kHz mono WAV) → `[{"label","score"}, ...]` (superb/wav2vec2-base-superb-er) - `GET /` → health Models load lazily on first use and stay warm. CPU-basic is enough — MMS is a 36M-param VITS and the classifier is a base wav2vec2.