Multi-voice Chinese short-drama dialogue with edge-tts (OpenMontage Drama Studio)

[charliewwdev]

I built OpenMontage Drama Studio, a self-hosted web app that turns a one-line idea into a short AI drama (短剧). edge-tts is the main voice engine, so I wanted to share how it's used.

What it does: a 7-stage pipeline: idea → script → cast → scenes → props → storyboard → video → voice → edit. You can review, edit and re-run each stage before moving on.

How edge-tts is used:

• It's the primary TTS backend (studio/server/audio_tts.py), called through edge_tts.Communicate. If the Python call fails, it falls back to the edge-tts CLI (--voice ... --write-media).
• Multi-voice dialogue: each speaker in the script gets a stable voice. Narration uses zh-CN-YunyangNeural by default, or EDGE_TTS_VOICE if set. Other characters are assigned from a female pool (zh-CN-XiaoxiaoNeural / XiaoyiNeural / XiaohanNeural) or a male pool (zh-CN-YunxiNeural / YunjianNeural / YunyangNeural), based on name hints plus a hash. That way two characters don't end up sounding like one narrator.
• The lines are synthesized one at a time and joined with ffmpeg, with a short gap between them. The result is optionally paired with a soft procedural background track that is also generated with ffmpeg, so no music API is needed.
• You can also pin a single voice, or switch to an OpenAI-compatible TTS endpoint instead, but edge-tts is the default because it needs no API key.

Run it: docker compose -f docker-compose.studio.yml up starts the API, web UI, Postgres and Redis. You bring S3/MinIO, an LLM key and a Volcengine Ark key (Seedream/Seedance are paid third-party APIs).

• Repo: https://github.com/charliewwdev/OpenMontage
• Demo: https://openmontage.dtok.io

It's AGPL-3.0. It's an independent product rebuilt from the calesthio/OpenMontage codebase and isn't affiliated with that project. It's early (v0.1.0), so feedback is welcome.