Fact · Speech recognition
30.60percent word error rate
NVIDIA Canary Qwen 2.5B had a word error rate of 30.60 percent on LibriSpeech test-clean with added white noise at -5 dB SNR.
Quote found word for word on the live source, last checked October 19, 2026
In the source's own words
“WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise | **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** | ... | 2.5.0 | Canary-Qwen-2.5B | 2.41% | 4.08% | 9.83% | 30.60% |”
canary-qwen-2.5b model card (raw README), NVIDIA, Hugging Face (NVIDIA model card), 2025, White-noise table, SNR -5 dB column.
Open the original- What it measures
- The card adds additive white noise to LibriSpeech test-clean at the stated signal-to-noise ratio (SNR) and reports the WER for each level.
- What it does not show
- Noise is added artificially to clean recordings, so it is not a recording made in a noisy room.
In the data
A row of Speech Recognition Accuracy: Word Error Rates, as Canary-Qwen 2.5B, noise -5 dB: 30.6 percent.
- System
- NVIDIA Canary Qwen 2.5B
- Kind
- Open model
- Test set
- LibriSpeech test-clean with white noise
- Language
- English
- Speech
- Noisy or distant
- Year
- 2025