AI Tools.

Search

automatic speech recognition by nvidia

parakeet-tdt-0.6b-v2

Parakeet-TDT-0.6B-v2 is NVIDIA's 600M-parameter English ASR model built on the FastConformer architecture with a Token-and-Duration Transducer (TDT) decoder. It was trained on the Granary and NeMo ASR 3.0 datasets and is listed on the Hugging Face ASR leaderboard. The NeMo framework is required for inference, which differs from the standard Transformers pipeline.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository nvidia/parakeet-tdt-0.6b-v2 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
nvidia
Pipeline tag
automatic-speech-recognition
Library
NeMo
Framework tags
PyTorch
License tag
cc-by-4.0 — read the license file in the repo before relying on it
Language tags
English (en)
Papers cited
arXiv:2305.05084, arXiv:2304.06795, arXiv:2406.00899, arXiv:2410.01036, arXiv:2505.13404
Datasets declared
nvidia/Granary, nvidia/nemo-asr-set-3.0
Downloads (HF counter at last fetch)
449,828
Likes (HF counter at last fetch)
1,534
Model card
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2

Use cases

  • High-throughput English speech transcription in production environments
  • Real-time transcription where low-latency TDT decoding is needed
  • Benchmarking ASR models on the Hugging Face ASR leaderboard
  • Integration into NVIDIA NeMo-based ASR pipelines
  • Long-form audio transcription with accurate word-level timestamps

Pros

  • FastConformer architecture delivers strong accuracy-to-compute ratios for English ASR
  • TDT decoder supports word-level duration prediction, enabling accurate timestamp generation
  • Trained on Granary and NeMo ASR 3.0, large-scale proprietary datasets curated by NVIDIA
  • Listed on the HF ASR leaderboard with eval-results metadata for objective comparison
  • CC-BY-4.0 license permits commercial use and redistribution with attribution
  • 1506 likes reflects strong community validation relative to similar ASR models

Cons

  • Requires NVIDIA NeMo framework — not directly usable with standard Transformers pipelines
  • English-only; no multilingual capability
  • 600M parameters is large relative to Whisper-small or medium for teams with tight inference budgets
  • NeMo dependency introduces a heavy installation footprint not suitable for lightweight deployments
  • Performance on heavily accented or non-native English speech may degrade without domain fine-tuning

Tags

nemoautomatic-speech-recognitionspeechaudioTransducerTDTFastConformerConformerpytorchNeMohf-asr-leaderboardendataset:nvidia/Granarydataset:nvidia/nemo-asr-set-3.0arxiv:2305.05084arxiv:2304.06795arxiv:2406.00899arxiv:2410.01036arxiv:2505.13404license:cc-by-4.0