ASR Model Checkpoints

This page lists all supported ASR model checkpoints released by NVIDIA NeMo. Benchmark scores for each model can be found on its HuggingFace model card.

Glossary

Term	Definition
ASR	Automatic Speech Recognition — transcribing speech to text
AST	Automatic Speech Translation — translating speech to text from one language to another
AED	Attention Encoder-Decoder — autoregressive decoder using cross-attention (Canary family)
CTC	Connectionist Temporal Classification — non-autoregressive decoder
RNN-T	Recurrent Neural Network Transducer — autoregressive streaming-friendly decoder
TDT	Token-and-Duration Transducer — extends RNN-T with duration prediction for faster inference
Hybrid	Joint RNN-T + CTC model — both decoders trained together, either usable at inference
PnC	Punctuation and Capitalization in the output
SALM	Speech Augmented Language Model — combines a speech encoder with a large language model
Streaming	Real-time / cache-aware inference capability
EU4	English, German, Spanish, French
EU25	English, German, Spanish, French, Italian, Polish, Portuguese, Dutch, Russian, Ukrainian, Belarusian, Croatian, Czech, Bulgarian, Danish, Estonian, Finnish, Greek, Hungarian, Latvian, Lithuanian, Maltese, Romanian, Slovak, Slovenian, Swedish

Canary Models (AED)

Multi-task encoder-decoder models supporting ASR, AST, PnC, and timestamps across multiple languages.

Model	Decoder	Capabilities	Language	Size
canary-1b-v2	AED	ASR, AST, PnC, timestamps	EU25	1B
canary-qwen-2.5b	SALM	ASR, AST, PnC, timestamps	EU25	2.5B
canary-1b-flash	AED	ASR, AST, PnC, timestamps, fast	EU4	1B
canary-180m-flash	AED	ASR, AST, PnC, timestamps, fast	EU4	180M
canary-1b	AED	ASR, AST, PnC	EU4	1B

Parakeet Models

High-accuracy ASR models built on the FastConformer encoder architecture. Parakeet, Nemotron Speech, and the stt_*_fastconformer_* models below all share the same underlying FastConformer encoder; the different names reflect release branding, not architectural differences.

Model	Decoder	Capabilities	Language	Size
parakeet-tdt-0.6b-v3	TDT	ASR, PnC, timestamps	English	0.6B
parakeet-tdt-0.6b-v2	TDT	ASR, PnC, timestamps	English	0.6B
parakeet-tdt-1.1b	TDT	ASR, timestamps	English	1.1B
parakeet-tdt_ctc-1.1b	Hybrid TDT+CTC	ASR, timestamps	English	1.1B
parakeet-tdt_ctc-0.6b-ja	Hybrid TDT+CTC	ASR, timestamps	Japanese	0.6B
parakeet-tdt_ctc-110m	Hybrid TDT+CTC	ASR, timestamps	English	110M
parakeet-rnnt-1.1b	RNN-T	ASR, timestamps	English	1.1B
parakeet-rnnt-0.6b	RNN-T	ASR, timestamps	English	0.6B
parakeet-ctc-1.1b	CTC	ASR	English	1.1B
parakeet-ctc-0.6b	CTC	ASR	English	0.6B
parakeet-ctc-0.6b-Vietnamese	CTC	ASR	Vietnamese	0.6B
parakeet-rnnt-110m-da-dk	RNN-T	ASR	Danish	110M

Streaming Models

Cache-aware models for real-time / low-latency inference.

Model	Decoder	Capabilities	Language	Size
nemotron-speech-streaming-en-0.6b	Hybrid	ASR, streaming	English	0.6B
multitalker-parakeet-streaming-0.6b-v1	RNN-T	ASR, multitalker, streaming	English	0.6B
parakeet_realtime_eou_120m-v1	RNN-T	ASR, end-of-utterance, streaming	English	120M
stt_en_fastconformer_hybrid_large_streaming_multi	Hybrid	ASR, streaming, multiple look-aheads	English	Large
stt_en_fastconformer_hybrid_medium_streaming_80ms_pc	Hybrid	ASR, PnC, streaming	English	Medium
stt_en_fastconformer_hybrid_medium_streaming_80ms	Hybrid	ASR, streaming	English	Medium
stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc	Hybrid	ASR, PnC, streaming	Georgian	Large
stt_en_fastconformer_hybrid_large_streaming_1040ms	Hybrid	ASR, streaming	English	Large

FastConformer English Models (Non-Streaming)

Model	Decoder	Capabilities	Language	Size
stt_en_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	English	Large
stt_en_fastconformer_ctc_large	CTC	ASR	English	Large
stt_en_fastconformer_ctc_xlarge	CTC	ASR	English	XLarge
stt_en_fastconformer_ctc_xxlarge	CTC	ASR	English	XXLarge
stt_en_fastconformer_transducer_large	RNN-T	ASR	English	Large
stt_en_fastconformer_transducer_xlarge	RNN-T	ASR	English	XLarge
stt_en_fastconformer_transducer_xxlarge	RNN-T	ASR	English	XXLarge
stt_en_fastconformer_tdt_large	TDT	ASR	English	Large

FastConformer Multilingual Models

Single-language FastConformer Hybrid models. Models with _pc suffix support punctuation and capitalization.

Model	Decoder	Capabilities	Language	Size
stt_multilingual_fastconformer_hybrid_large_pc_blend_eu	Hybrid	ASR, PnC	Multilingual EU	Large
stt_de_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	German	Large
stt_es_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Spanish	Large
stt_es_fastconformer_hybrid_large_pc_nc	Hybrid	ASR, Punctuation only	Spanish	Large
stt_fr_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	French	Large
stt_it_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Italian	Large
stt_ru_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Russian	Large
stt_ua_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Ukrainian	Large
stt_pl_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Polish	Large
stt_hr_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Croatian	Large
stt_be_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Belarusian	Large
stt_nl_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Dutch	Large
stt_pt_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Portuguese	Large
stt_fa_fastconformer_hybrid_large	Hybrid	ASR	Farsi	Large
stt_ka_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Georgian	Large
stt_hy_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Armenian	Large
stt_ar_fastconformer_hybrid_large_pc_v1.0	Hybrid	ASR, PnC	Arabic	Large
stt_ar_fastconformer_hybrid_large_pcd_v1.0	Hybrid	ASR, PnC (diacritized)	Arabic	Large
stt_uz_fastconformer_hybrid_large_pc	Hybrid	ASR, PnC	Uzbek	Large
stt_kk_ru_fastconformer_hybrid_large	Hybrid	ASR	Kazakh + Russian	Large

Loading Models

All models (except SALM — see SpeechLM2) can be loaded via the from_pretrained() API:

import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.from_pretrained("nvidia/parakeet-tdt-0.6b-v2")