[KongchangAI]
Model

Canary

An automatic speech recognition model in the NVIDIA Nemotron speech model family. It supports multilingual streaming and batch transcription.

Timeline (last 90 days)

Oct 23

Unsloth reduces training memory for the speech models Parakeet, Canary, and CTC by up to 55%.

Unverified50%
Oct 4

Progressive delivery uses a canary approach to gradually increase traffic—first to 1% or 2% of users, then more after stability is confirmed, with an immediate rollback to the previous stable version (N-1) upon detecting anomalies.

Unverified50%
Sep 29

NVIDIA's Nemotron speech model series includes Parakeet, Canary, and Nemotron ASR in the recognition direction, collectively supporting streaming and batch transcription in 43 languages.

Unverified50%

All Facts (3)

Source Articles