# xzf-thu/Mega-ASR

First foundation ASR built for the real world - 7 atomic acoustic conditions, 54 compound scenarios, 2.6M samples, and up to ~30% gains over SOTA where every other model falls apart. **You'll come back to MEGA-ASR, after the rest fail in the wild. ⭐**

Repository: https://github.com/xzf-thu/Mega-ASR
Canonical: https://ross.abutalabs.com/products/mega-asr
Homepage: https://xzf-thu.github.io/Mega-ASR/
Language: Python
License Family: other
Topics: asr, robust
Last push: 2026-09-02T06:19:55+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 35, longevity 7
- inputs: {"age_days": 109, "days_push": 0, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1136, forks 74 (observed 2026-09-03T02:15:20.754553+00:00)

## What it is
Mega-ASR is a foundation automatic speech recognition model trained on 2.6M samples spanning 7 atomic acoustic conditions and 54 compound real-world scenarios, using A2S-SFT and DG-WGPO reinforcement learning. The repository provides training code, model weights, and an associated dataset and benchmark for robust in-the-wild speech recognition.

## Use cases
- transcribe speech in noisy real-world environments
- recognize far-field or reverberant audio
- evaluate ASR robustness across acoustic conditions
- train a robust speech recognition model
- benchmark ASR models against challenging audio
- download open ASR model weights

## When to choose
- your ASR pipeline fails on noisy, far-field, or distorted real-world audio
- you need a robust open ASR foundation model with weights and training data
- you want to benchmark speech recognition under compound acoustic scenarios

## When to avoid
- you need a lightweight ASR for clean studio audio
- you require a permissive license - the repo has no license
- you need streaming or low-latency on-device transcription

## Facets
- artifact type: library
- maturity: active
- function: speech-recognition, machine-learning, deep-learning, llm-training, data-generation
- domain: speech-processing, machine-learning, artificial-intelligence
- platform: python
- tags: asr, robust-speech-recognition, acoustic-simulation, foundation-model, reinforcement-learning, dataset, model-weights, audio, gpu, linux

## Member repositories
- xzf-thu/Mega-ASR (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:20.754553+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:37:54.945657+00:00, confidence not recorded.
  - readme: https://github.com/xzf-thu/Mega-ASR (fetched 2026-09-03T02:15:20.754553+00:00, sha 1b4960e365cf)
  - homepage: https://xzf-thu.github.io/Mega-ASR/ (fetched 2026-08-29T12:42:23.727632+00:00, sha 0846541fa75a)
- Data as of 2026-08-30T08:39:29.467469+00:00.