#8989th of 93 in VoiceFor voice, speech and music

soulx-podcast

Multi-speaker podcast speech synthesis with dialects and voice cloning

Stars
3.7k
License
Apache-2.0
Last commit
Dec 2025
Language
Python
Stars, last 29 weeks+399
25 Apr 2026: 3,272 stars2 May 2026: 3,283 stars9 May 2026: 3,299 stars16 May 2026: 3,312 stars23 May 2026: 3,332 stars30 May 2026: 3,369 stars6 Jun 2026: 3,402 stars13 Jun 2026: 3,423 stars20 Jun 2026: 3,440 stars27 Jun 2026: 3,457 stars4 Jul 2026: 3,466 stars11 Jul 2026: 3,480 stars18 Jul 2026: 3,490 stars25 Jul 2026: 3,495 stars1 Aug 2026: 3,506 stars8 Aug 2026: 3,512 stars15 Aug 2026: 3,521 stars22 Aug 2026: 3,530 stars29 Aug 2026: 3,540 stars5 Sept 2026: 3,549 stars12 Sept 2026: 3,554 stars19 Sept 2026: 3,563 stars26 Sept 2026: 3,568 stars3 Oct 2026: 3,589 stars10 Oct 2026: 3,608 stars17 Oct 2026: 3,625 stars24 Oct 2026: 3,647 stars31 Oct 2026: 3,658 stars7 Nov 2026: 3,663 stars14 Nov 2026: 3,671 stars

3,272 on 25 Apr 20263,671 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Inference code and 1.7B checkpoints for generating multi-turn, multi-speaker dialogue speech with zero-shot voice cloning. It supports Mandarin and English, plus Sichuanese, Henanese and Cantonese through a separate dialect model. Paralinguistic tags such as laughter, sigh and breathing can be inserted. A WebUI and an optional vLLM Docker runtime are included.

Who it is for: Developers generating multi-speaker podcast audio, including Chinese dialects

Strengths

  • Generates long-form dialogue with multiple speakers, not only single-voice TTS
  • Zero-shot voice cloning works across Chinese dialects from Mandarin prompts
  • Paralinguistic tags: laughter, sigh, breathing, coughing, throat clearing
  • Apache-2.0 code and weights; WebUI and optional vLLM Docker runtime provided

Weaknesses

  • Streaming inference is not implemented yet (listed in TODOs)
  • Dialects need the separate dialect checkpoint rather than the base model
  • Languages limited to Mandarin, English and three Chinese dialects
  • No tagged releases; install is a manual conda and pip setup

What it needs

  • Needs Python 3.11, Conda, Hugging Face model weights, Docker with NVIDIA runtime (optional vLLM)
  • Models: SoulX-Podcast-1.7B, SoulX-Podcast-1.7B-dialect
  • port 7860

Also in Voice

See all 93
Also in Voice
RankProjectScore
1voicestudioDesktop app for voice cloning, dubbing, dictation and audiobooks113k starsAGPL-3.098 out of 100
2handyOffline push-to-talk dictation app that pastes transcripts into any text field39.1k starsMIT92 out of 100
3whisper.cppC/C++ speech-to-text engine running OpenAI Whisper models locally55.3k starsMIT90 out of 100
#4meetilyDesktop app that records, transcribes and summarizes meetings locally Live demo ↗ (opens in a new tab)32.8k starsMIT87 out of 100
#5buzzOffline desktop app for transcribing and translating audio with Whisper22.9k starsMIT84 out of 100