[Code] [Paper] [Model] [Blog] [Demo]
FireRedASR2-LLM is the 8B+ parameter variant of the FireRedASR2 system, designed to achieve state-of-the-art performance and enable seamless end-to-end speech interaction. It adopts an Encoder-Adapter-LLM framework leveraging large language model capabilities.
The model was introduced in the paper FireRedASR2S: A State-of-the-Art Industrial-Grade All-in-One Automatic Speech Recognition System.
Authors: Kaituo Xu, Yan Jia, Kai Huang, Junjie Chen, Wenpeng Li, Kun Liu, Feng-Long Xie, Xu Tang, Yao Hu.
🔥 News
- [2026.03.12] 🔥 We release FireRedASR2S technical report. See arXiv.
- [2026.03.05] 🚀 vLLM supports FireRedASR2-LLM.
- [2026.02.25] 🔥 We release FireRedASR2-LLM model weights. 🤗 🤖
Sample Usage
To use this model, please refer to the installation and setup instructions in the official GitHub repository.
from fireredasr2s.fireredasr2 import FireRedAsr2, FireRedAsr2Config
batch_uttid = ["hello_zh", "hello_en"]
batch_wav_path = ["assets/hello_zh.wav", "assets/hello_en.wav"]
# FireRedASR2-LLM Configuration
asr_config = FireRedAsr2Config(
use_gpu=True,
decode_min_len=0,
repetition_penalty=1.0,
llm_length_penalty=0.0,
temperature=1.0
)
# Load the model
model = FireRedAsr2.from_pretrained("llm", "FireRedTeam/FireRedASR2-LLM", asr_config)
# Transcribe
results = model.transcribe(batch_uttid, batch_wav_path)
print(results)
# [{'uttid': 'hello_zh', 'text': '你好世界', 'rtf': '0.0681', 'wav': 'assets/hello_zh.wav'}, {'uttid': 'hello_en', 'text': 'hello speech', 'rtf': '0.0681', 'wav': 'assets/hello_en.wav'}]
Evaluation
FireRedASR2-LLM achieves state-of-the-art accuracy across Mandarin and various Chinese dialects.
| Metric | FireRedASR2-LLM | Doubao-ASR | Qwen3-ASR | Fun-ASR |
|---|---|---|---|---|
| Avg CER (Mandarin, 4 sets) | 2.89 | 3.69 | 3.76 | 4.16 |
| Avg CER (Dialects, 19 sets) | 11.55 | 15.39 | 11.85 | 12.76 |
FAQ
Q: What audio format is supported?
16kHz 16-bit mono PCM wav. You can convert files using ffmpeg:
ffmpeg -i <input_audio_path> -ar 16000 -ac 1 -acodec pcm_s16le -f wav <output_wav_path>
Q: What are the input length limitations? FireRedASR2-LLM supports audio input up to 40s.
Citation
@article{xu2026fireredasr2s,
title={FireRedASR2S: A State-of-the-Art Industrial-Grade All-in-One Automatic Speech Recognition System},
author={Xu, Kaituo and Jia, Yan and Huang, Kai and Chen, Junjie and Li, Wenpeng and Liu, Kun and Xie, Feng-Long and Tang, Xu and Hu, Yao},
journal={arXiv preprint arXiv:2603.10420},
year={2026}
}