Canary-1B
Multilingual ASR plus speech translation in 25 languages — commercial-friendly open weights
About
Canary is NVIDIA NeMo's multilingual speech model: a 978M-parameter FastConformer encoder with a transformer decoder that does both transcription and speech translation across 25 European languages, including X-to-English and English-to-X directions. Version 2 is the current checkpoint on Hugging Face (nvidia/canary-1b-v2), and NVIDIA documents it alongside Parakeet in a joint technical report. The weights are CC-BY-4.0, which permits commercial use with attribution — a real differentiator next to non-commercial research models. The main model does not emit word-level timestamps itself; NVIDIA ships an auxiliary CTC aligner inside the checkpoint for that. There is also no vocabulary biasing or hotword support, so proper nouns outside the training vocabulary can suffer.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: results documented in the Canary-1B-v2 and Parakeet-TDT-0.6B-v3 technical report (arXiv 2509.14128)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: HF nvidia/canary-1b-v2 and CC-BY-4.0 commercial-use terms verified; 978M params, 25 languages, X-to-En/En-to-X translation capability verified; Exact v2 release day not verified — releasedAt null.