nemotron-asr-streaming Model by NVIDIA

Model Overview

Description

This NIM contains 2 profiles:

nvidia/nemotron-speech-streaming-en-0.6b (Nemotron 3 ASR - English)
nvidia/nemotron-3.5-asr-streaming-0.6b (Nemotron 3.5 ASR - 40 language-locales)

Nemotron 3 ASR is a 600-million-parameter streaming Automatic Speech Recognition (ASR) model engineered for high-quality English transcription across both low-latency streaming and high-throughput batch workloads. Developed by NVIDIA, it natively outputs punctuation and capitalization with no post-processing required.

By leveraging a Cache-Aware FastConformer-RNNT architecture, the model processes only new audio chunks while reusing cached encoder context - eliminating redundant overlapping computations and minimizing end-to-end latency without sacrificing accuracy. Trained on over 250,000 hours of US English (en-US) speech, it performs robustly across diverse acoustic conditions.

Key Features:

Low-latency streaming with zero overlapping computations
Native punctuation and capitalization output
250,000+ hours of en-US training data for strong noise robustness
Supports both streaming and high-throughput batch inference

Nemotron 3.5 ASR extends Nemotron 3 ASR from English to 40 languages and locales, with automatic language detection - no language selection required at runtime. Built on the same Cache-Aware FastConformer-RNNT architecture, it is designed for the same low-latency streaming use cases as its predecessor, now serving global, multilingual deployments.

Key Features:

40 supported languages and locales with automatic language detection
Low-latency streaming with zero overlapping computations
Native punctuation and capitalization output
Drop-in multilingual replacement for Nemotron 3 ASR streaming pipelines

License/Terms of Use

GOVERNING TERMS: This trial service is governed by the NVIDIA API Trial Terms of Service. The NIM container is governed by the NVIDIA Software License Agreement; use of the model is governed by the NVIDIA Open Model License Agreement.

{note}

To deploy the multilingual variant of this model, Nemotron 3.5 ASR, with 40 language-locales, set
`NIM_TAGS_SELECTOR=type=multi`

The model includes automatic language detection and is intended for the same streaming use cases as Nemotron 3 ASR which is English only. Refer to the deployment instructions for usage details.

Deployment Selectors

Please see the Deploy tab for reference.

Profile	`NIM_TAGS_SELECTOR`	HuggingFace
Nemotron 3 ASR (English)	(default is en-US - for English, no selector needed)	nvidia/nemotron-speech-streaming-en-0.6b
Nemotron 3.5 ASR (40 languages)	`type=multi`	nvidia/nemotron-3.5-asr-streaming-0.6b

Release Date

Nemotron 3 ASR: 03/12/2026

build.nvidia.com: build.nvidia.com/nvidia/nemotron-asr-streaming
Hugging Face: nvidia/nemotron-speech-streaming-en-0.6b
NGC: catalog.ngc.nvidia.com/orgs/nim/nvidia/containers/nemotron-asr-streaming

Nemotron 3.5 ASR: June 4 and June 11, 2026

build.nvidia.com: build.nvidia.com/nvidia/nemotron-asr-streaming
Hugging Face: nvidia/nemotron-3.5-asr-streaming-0.6b
NGC: catalog.ngc.nvidia.com/orgs/nim/nvidia/containers/nemotron-asr-streaming