3.6K ·GitHub
What is this?
SoulX-Podcast is an inference codebase that generates realistic, long-form, multi-speaker podcast-style dialogue from text, including paralinguistic effects and several Chinese dialects.
SoulX-Podcast is an inference codebase that generates realistic, long-form, multi-speaker podcast-style dialogue from text, including paralinguistic effects and several Chinese dialects.Is this a ready-made app or code I run myself?It is a research codebase, not a packaged app: you clone the repo, download the model checkpoints, and run the provided scripts or a WebUI yourself.
It is a research codebase, not a packaged app: you clone the repo, download the model checkpoints, and run the provided scripts or a WebUI yourself.How do I get it running?Clone the repository, install Conda, create the Conda environment, then download the model checkpoints from Hugging Face (via Python script or git clone) before running the demo or WebUI scripts.
Clone the repository, install Conda, create the Conda environment, then download the model checkpoints from Hugging Face (via Python script or git clone) before running the demo or WebUI scripts.What do I need before installing?You need Conda, a Linux machine, and the SoulX-Podcast model checkpoints from Hugging Face; an optional modified build of vLLM 0.10.1 can be used for faster inference.
You need Conda, a Linux machine, and the SoulX-Podcast model checkpoints from Hugging Face; an optional modified build of vLLM 0.10.1 can be used for faster inference.Any install pitfalls or missing features right now?The install instructions only cover Linux, and streaming inference is still listed as an open TODO rather than something available today.
The install instructions only cover Linux, and streaming inference is still listed as an open TODO rather than something available today.Who is this meant for?It targets researchers and developers working on podcast-style TTS and voice cloning, with use restricted to academic, educational, and other legitimate applications per the project's disclaimer.
It targets researchers and developers working on podcast-style TTS and voice cloning, with use restricted to academic, educational, and other legitimate applications per the project's disclaimer.What are the license terms?The code is released under Apache 2.0, but the authors explicitly prohibit misuse such as unauthorized voice cloning, impersonation, fraud, or deepfakes.
The code is released under Apache 2.0, but the authors explicitly prohibit misuse such as unauthorized voice cloning, impersonation, fraud, or deepfakes.What dialects does it support besides Mandarin and English?It supports several Chinese dialects, including Sichuanese, Henanese, and Cantonese, for zero-shot voice cloning and speech generation.
It supports several Chinese dialects, including Sichuanese, Henanese, and Cantonese, for zero-shot voice cloning and speech generation.Can it add natural effects like laughter to the speech?Yes, it includes paralinguistic tags for effects such as laughter and sighs to make generated dialogue sound more realistic.
Yes, it includes paralinguistic tags for effects such as laughter and sighs to make generated dialogue sound more realistic.Is there a web interface, or only command-line scripts?There is a WebUI for easy inference in addition to the command-line demo scripts for monologue TTS.
There is a WebUI for easy inference in addition to the command-line demo scripts for monologue TTS.Can inference be accelerated or containerized?Yes, the project ships a modified vLLM 0.10.1 build for acceleration and, as of November 2025, also supports running via Docker with vLLM.
Yes, the project ships a modified vLLM 0.10.1 build for acceleration and, as of November 2025, also supports running via Docker with vLLM.What are similar projects?VoxCPM, ChatTTS, Index-Translate, fish-speech, MOSS-TTS, MOSS-TTS-Nano, FireRedTTS3, Speech
VoxCPM, ChatTTS, Index-Translate, fish-speech, MOSS-TTS, MOSS-TTS-Nano, FireRedTTS3, SpeechWhich topics is it in?podcast, speech, speech-synthesis, webui, tts, vllm, text-to-speech, academic-research, multilingual, audio, inference, huggingface
podcast, speech, speech-synthesis, webui, tts, vllm, text-to-speech, academic-research, multilingual, audio, inference, huggingface