SoulX-Podcast
Speech model for stable 90-minute multi-speaker podcasts — the hard problem of long audio
A speech model that stably generates 90-minute multi-speaker podcast episodes — the hard problem of long-form audio generation.
Voice AI: lifelike TTS, accurate ASR, and real-time voice agents.
Key features
- Long-form multi-turn multi-speaker dialogic speech generation
- Cross-dialectal zero-shot voice cloning (Sichuanese, Henanese, Cantonese)
- Paralinguistic controls: laughter, sighs, breathing, coughing tags
- Also performs well on conventional monologue TTS
Good for
- Podcast creators generating multi-speaker episodes from scripts
- Developers building Chinese-dialect voice content
Stack & tags
PythonvLLMpodcastttsdialoguevoice-cloningchinese-dialectsparalinguistic
Quick start
pip install -r requirements.txt
bash example/infer_dialogue.shFrom the project README on GitHub.