# Soul-AILab releases SoulX-Podcast-1.7B, a multi-speaker podcast TTS in English and Chinese

> SoulX-Podcast-1.7B generates multi-turn, multi-speaker dialogue with cross-dialect cloning in English and Chinese; Apache 2.0.

Published 2025-10-27

Canonical: https://shipped.anth.us/open-weights/soul-ailab-soulx-podcast-1-7b/

Soul-AILab published SoulX-Podcast-1.7B, a text-to-speech model designed for podcast-style multi-turn, multi-speaker dialogue, which also handles conventional monologue. The Hub lists it at 2B parameters in BF16 though the name says 1.7B. It supports English and Chinese plus the Sichuanese, Henanese and Cantonese dialects, the dialect inference using a separate SoulX-Podcast-1.7B-dialect model. Features listed are long-form multi-speaker generation, cross-dialectal zero-shot voice cloning, and paralinguistic controls such as laughter and sighs.

The card gives no numeric benchmarks, only a radar chart image. Its TODO list shows monologue example scripts, a WebUI, a Spaces demo, vLLM or Docker support and streaming inference as not yet done. No Inference Provider hosts it.

The licence is Apache 2.0. A disclaimer prohibits unauthorised voice cloning, impersonation, fraud and deepfakes. The card gives no release date; the date used is the Hugging Face repo creation date, 27 October 2025, which matches the linked paper's listed publication date.

A model for a specific job, with the tooling still on the to-do list.

## Sources
- [Soul-AILab model card on Hugging Face](https://huggingface.co/Soul-AILab/SoulX-Podcast-1.7B)
- [Hugging Face repo metadata (creation date)](https://huggingface.co/api/models/Soul-AILab/SoulX-Podcast-1.7B)

## Editor's first pass (AI-assisted)
- Real 3.5: Apache 2.0 weights, but the card's TODO list shows the WebUI, vLLM and streaming still undone.
- Useful 3.5: Podcast-style dialogue with dialect cloning is a real use case; no numbers on the card.
- Notable 3: A modest following and a niche focus.
- Novel 3.5: Dialogue-first design with Sichuanese, Henanese and Cantonese support.