FunASR

@MODELSCOPE

combines real-time transcription, sentence segmentation, speaker differentiation, and speech services into a complete auditory toolbox.

FunASR not only dictates text, but also provides real-time streaming, sentence segmentation, punctuation and speaker pipelines, making it suitable as a complete auditory infrastructure for Chinese companion scenes.

Project address (can be copied to AI):https://github.com/modelscope/FunASR

PROJECT INTRO

Project introduction

FunASR provides speech recognition training and inference, streaming ASR, speech activity detection, punctuation recovery, and speaker separation, and can be packaged as an OpenAI-compatible or MCP service.

The companion system can use it to handle real-time microphones, voice messages, and transcription in multi-person environments, and then pass the text with time and speaker information to the conversation model.

Suitable for:Projects that require a Chinese-friendly, self-hosted complete speech recognition pipeline.
Platform:Python, a self-hosted CPU/GPU environment.
Usage:Select models by task, combining ASR, VAD, punctuation and speaker modules.
License:Toolkit code is MIT; pre-training weight permission is confirmed separately for each model card.

◌

COMMENTS & FEEDBACK

Comments and Feedback

Guest message · Will be published after review

People who have used this project can come back and tell newcomers: which platform it ran on, whether there were any pitfalls during the installation, and what the actual experience was like. The current version is in guest mode, and you do not need to register an account to leave a message.

Reading public comments...

SOURCE & CREDIT

Atlas is responsible for the introduction and navigation, and the use still returns to the original author.

We do not mirror project files, nor do we intercept author traffic. For installation, download, version updates and the latest instructions, please refer to the project page provided by the original author.

Project link ↗