OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Soulx Transcriber

Soulx Transcriber is an end-to-end Python framework that transcribes multi-speaker audio while assigning speaker labels and timestamps to recognized speech.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Soulx Transcriber — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when a multi-speaker recording needs both transcript content and structured speaker and timing information, particularly for conversations with overlapping or rapid turns.

Who it is for

  • Developers building transcription workflows for meetings or group discussions.
  • Researchers and practitioners working with speaker-attributed interview transcripts.
  • Teams needing time-aligned transcripts from multi-speaker audio.

Documented uses

  • Convert a WAV recording into recognized speech with speaker attribution and timestamps.
  • Create transcripts that make speaker turns easier to follow.
  • Explore continued pre-training or fine-tuning through MS-Swift.
Before you choose

Requirements & limits

The project is public source code under the Apache-2.0 license. Its README links pretrained weights on Hugging Face and ModelScope.

  • Inference instructions recommend vLLM-omni and require a separate environment setup beyond installing the project dependencies.
  • The project builds on Qwen3-Omni-30B-A3B-Instruct.
  • The README describes continued pre-training and fine-tuning with MS-Swift; it says training scripts are forthcoming.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the environment and run WAV inference

Set up the documented vLLM-omni inference environment and start the provided single-WAV inference script.

Before you start

  • Python 3.12
  • The SoulX-Transcriber repository
  • uv, vLLM and vLLM-omni as specified in the repository instructions
  • A WAV file for the documented single-file inference workflow

Steps

  1. Clone the SoulX-Transcriber repository and create and activate a Python 3.12 environment.

  2. Install uv, create and activate a vLLM-omni environment, then install vLLM and vLLM-omni.

  3. Clone the vLLM-omni repository and install it in editable mode as shown in the instructions.

  4. Download the pretrained model, activate the inference environment and run bash ./inference.sh.

Expected outcome

The documented workflow starts inference for a single WAV file using the pretrained model.

Guide limitations

  • The README provides commands but no hardware requirements or resource estimates.
  • It does not document the inference output format in the supplied instructions.
  • The repository notes an editable source installation as a possible response to an undefined-symbol error with VLLM_USE_PRECOMPILED=1.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
Soul-AILab/SoulX-Transcriber · Official homepage title · retained in its source language
How we review sourcesSuggest a correction