NVIDIANemotron 3.5
Nemotron 3.5 ASR Streaming 0.6B
This 600M-parameter speech recognizer supports streaming transcription across 40 language-locales.
What the release provides
Its cache-aware FastConformer-RNNT architecture reuses prior audio state. Chunk sizes range from 80 to 1,120 milliseconds. Language-ID prompting or automatic language detection controls the language path.Nemotron 3.5 ASR Streaming 0.6B official model card
How to use it well
Practical guidance. Choose a chunk size with the complete audio pipeline running. Smaller chunks can improve responsiveness but change the work done per second. Review partial transcript revisions and the final utterance separately, especially when another system acts on the text.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nvidia/nemotron-3.5-asr-streaming-0.6b |