Soul App Open-Sources SoulX-Podcast: A Breakthrough in Multi-Speaker Podcast Voice Synthesis

Soul App’s AI team (Soul AI Lab) has officially open-sourced SoulX-Podcast, a podcast-specific voice synthesis model optimized for multi-speaker, multi-turn conversations. The full release includes a live demo, technical report, source code, and Hugging Face resources, empowering developers with end-to-end support.

Designed for podcast production, SoulX-Podcast excels in:

• Long-form fluency: Stable generation of 60+ minute dialogues with accurate speaker transitions and natural prosody.
• Paralinguistic realism: Includes laughter, throat-clearing, and expressive nuances for immersive audio.
• Multilingual & dialect support: Beyond Mandarin and English, it generates Sichuanese, Henanese, Cantonese, and enables cross-dialect cloning using standard Mandarin references.
• Zero-shot voice cloning: Replicates speaker style from minimal audio, dynamically adjusting rhythm based on context.

The open-source move aligns with Soul’s “AI + Social” strategy. Known for voice-first social features—like full-duplex AI calls and virtual hosts “Meng Zhishi” and “Yuni,” which powered a 40-minute group chat party in September—Soul identified a gap in open-source podcast TTS. By releasing SoulX-Podcast, the team aims to collaborate with the AIGC community to advance voice tech in content creation and virtual interaction.

Soul AI Lab pledges ongoing improvements in conversational synthesis and human-like expression, deepening open-source contributions to deliver warmer, more engaging AI social experiences.

Resources Demo Page：https://soul-ailab.github.io/soulx-podcast Technical Report：https://arxiv.org/pdf/2510.23541 Source Code：https://github.com/Soul-AILab/SoulX-Podcast HuggingFace：https://huggingface.co/collections/Soul-AILab/soulx-podcast

Source: Soul AI Lab