antirez

antirez / qwen-asr

Public

C inference for Qwen3-ASR 0.6b and 1.7b transcriptions models

437
41
100% credibility
Found Feb 12, 2026 at 105 stars 4x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C
AI Summary

A lightweight C program for running Qwen3-ASR speech-to-text models on CPU with support for both offline and streaming transcription.

How It Works

1
🔍 Find a fast speech tool

You discover a simple tool that turns spoken words into text quickly on your computer.

2
🔨 Set it up easily

You build the tool with one easy command, no complicated setups needed.

3
📥 Grab a model

Choose a small or large brain for the tool and download it with a quick script.

4
🎤 Add your audio

Point it at any sound file like a recording or podcast, or pipe in live audio.

5
✨ See text appear

Words stream out live as the tool listens and writes them down accurately.

6
🔄 Try streaming mode

Switch to live mode for ongoing talks, keeping everything smooth and continuous.

✅ Perfect transcripts

You get clear, fast text from any audio, ready to use or share right away.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 105 to 437 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is qwen-asr?

Pure C inference for Qwen3-ASR speech-to-text models (0.6B and 1.7B), turning WAV files or stdin audio into streaming text transcripts. Pipe any format via ffmpeg, force languages like Italian, or bias with prompts for terms like PostgreSQL. Delivers realtime multiples on CPU hardware, with offline full-file or chunked streaming modes.

Why is it gaining traction?

Minimal deps (just BLAS), mmap'd weights for instant loads, and CPU speeds crushing Whisper alternatives—7-13x realtime on M3 Max. Streaming handles live audio with prefix rollback for stable incremental output, perfect for qwen asr realtime without GPU. Antirez pedigree makes it a go-to for github inference models seeking qwen asr vs whisper edges.

Who should use this?

Embedded engineers on Jetson or servers needing qwen asr api without Python bloat. Devs scripting qwen asr toolkit pipelines via CLI/stdin, or integrating C API into low-latency apps like meeting transcribers. Suited for qwen 3 asr github fans prioritizing CPU github llm inference over flashier stacks.

Verdict

Solid for 0.6B CPU transcription; benchmarks and regression tests impress despite 72 stars and 1.0% credibility. Early but battle-tested by Redis author—fork if needed, watch for polish.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.