Supported Languages:
Bulgarian (bg), Croatian (hr), Czech (cs), Danish (da), Dutch (nl), English (en), Estonian (et), Finnish (fi), French (fr), German (de), Greek (el), Hungarian (hu), Italian (it), Latvian (lv), Lithuanian (lt), Maltese (mt), Polish (pl), Portuguese (pt), Romanian (ro), Slovak (sk), Slovenian (sl), Spanish (es), Swedish (sv), Russian (ru), Ukrainian (uk)

🗣️ Experience Canary-1b-v2 in action at Hugging Face Demo

Canary-1b-v2 model is ready for commercial/non-commercial use.

License/Terms of Use:

GOVERNING TERMS: Use of this model is governed by the CC-BY-4.0 license.

Discover more from NVIDIA:

For documentation, deployment guides, enterprise-ready APIs, and the latest open models—including Nemotron and other cutting-edge speech, translation, and generative AI—visit the NVIDIA Developer Portal at developer.nvidia.com. Join the community to access tools, support, and resources to accelerate your development with NVIDIA’s NeMo, Riva, NIM, and foundation models.

Explore more from NVIDIA:

What is Nemotron?
NVIDIA Developer Nemotron
NVIDIA Riva Speech
NeMo Documentation

Key Features

Canary-1b-v2 is a scaled and enhanced version of the Canary model family, offering:

Support for 25 European languages, expanding from the 4 languages in canary-1b/canary-1b-flash to 21 additional languages
State-of-the-art performance among models of similar size
Comparable quality to models 3× larger, while being up to 10× faster
Automatic punctuation and capitalization
Accurate word-level and segment-level timestamps
Segment-level timestamps also available for translated outputs
Released under a permissive CC BY 4.0 license

Canary-1b-v2 model is the first model from NeMo team that leveraged full Nvidia's Granary dataset [1] [2], showcasing its multitask and multilingual capabilities.

For full details on the model architecture, training methodology, datasets, and evaluation results, check out the Canary-1b-v2 Technical Report.

For a deeper glimpse into the Canary family of models, explore this comprehensive NeMo tutorial on multitask speech models.

Automatic Speech Recognition (ASR)

Figure 1: ASR WER comparison across different models. This does not include Punctuation and Capitalisation errors.

Speech Translation (AST)

X → English

Figure 2: AST X → En COMET scores comparison across different models

English → X

Figure 3: AST En → X COMET scores comparison across different models

Evaluation Notes

Note 1: The above evaluations are conducted in two settings: (1) All supported languages (24 languages, excluding Latvian since seamless-m4t-v2-large and seamless-m4t-medium do not support it), and (2) Common languages (6 languages supported by all compared models: en, fr, de, it, pt, es).

Note 2: Performance differences may be partly attributed to Portuguese variant differences - our training data uses European Portuguese while most benchmarks use Brazilian Portuguese.

Deployment Geography

Global

Use case

This model serves developers, researchers, academics, and industries building applications that require speech-to-text capabilities, including but not limited to: conversational AI, voice assistants, transcription services, subtitle generation, and voice analytics platforms.

Release Date

Huggingface 08/14/2025

Model Architecture

Canary-1b-v2 is an encoder-decoder architecture featuring a FastConformer Encoder [3] and a Transformer Decoder [4]. The model extracts audio features through the encoder and uses task-specific tokens—such as <source language> and <target language>—to guide the Transformer Decoder in generating text output.

It uses a unified SentencePiece Tokenizer [5] with a vocabulary of 16,384 tokens, optimized across all 25 supported languages. The architecture includes 32 encoder layers and 8 decoder layers, totaling 978 million parameters.

For implementation details, see the NeMo repository.

Input

Input Type(s): 16kHz Audio
Input Format(s): .wav and .flac audio formats
Input Parameters: 1D (audio signal)
Other Properties Related to Input: Monochannel audio

Output

Output Type(s): Text
Output Format: String
Output Parameters: 1D (text)
Other Properties Related to Output: Punctuation and Capitalization included.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

How to Use This Model

To train, fine-tune or play with the model you will need to install NVIDIA NeMo [6]. We recommend you install it after you've installed latest PyTorch version.

pip install -U nemo_toolkit['asr']

The model is available for use in the NeMo toolkit [6], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

You can also run Canary with Transformers 🤗 (more below).

1) NeMo usage

Automatically instantiate the model

python

from nemo.collections.asr.models import ASRModel
asr_ast_model = ASRModel.from_pretrained(model_name="nvidia/canary-1b-v2")

Transcribing using Python

First, let's get a sample:

wget https://dldata-public.s3.us-east-2.amazonaws.com/2086-149220-0033.wav

Then simply do:

python

output = asr_ast_model.transcribe(['2086-149220-0033.wav'], source_lang='en', target_lang='en')
print(output[0].text)

Translating using Python

Be sure to specify necessary target_lang for proper translation:

python

output = asr_ast_model.transcribe(['2086-149220-0033.wav'], source_lang='en', target_lang='fr')
print(output[0].text)

Transcribing with timestamps

Note: Use main branch of NeMo to get timestamps until it is released in NeMo 2.5.

To transcribe with timestamps:

python

output = asr_model.transcribe(['2086-149220-0033.wav'], source_lang='en', target_lang='en', timestamps=True)
# by default, timestamps are enabled for word and segment level
word_timestamps = output[0].timestamp['word'] # word level timestamps for first sample
segment_timestamps = output[0].timestamp['segment'] # segment level timestamps

for stamp in segment_timestamps:
    print(f"{stamp['start']}s - {stamp['end']}s : {stamp['segment']}")

Translating with timestamps

To translate with timestamps:

python

output = asr_model.transcribe(['2086-149220-0033.wav'], source_lang='en', target_lang='fr', timestamps=True)

segment_timestamps = output[0].timestamp['segment'] # only supports segment level timestamps for translation

for stamp in segment_timestamps:
    print(f"{stamp['start']}s - {stamp['end']}s : {stamp['segment']}")

For translation task, please, refer to segment-level timestamps for getting intuitive and accurate alignment.

Note: If timestamps are not required for your work, you can reduce memory usage by restoring only the .nemo file without the auxiliary CTC model. To do this, extract the .nemo file, remove any timestamps_asr_model files, then repackage it into a new .nemo file.

2) Transformers 🤗 usage

Until Canary is part of an official Transformers release, you can use it by installing from source.

pip install git+https://github.com/huggingface/transformers

➡️ Pipeline usage

python

from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", model="nvidia/canary-1b-v2")
out = pipe("https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/bcn_weather.mp3")
print(out)

➡️ Transcription

python

from datasets import load_dataset, Audio
from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq

model_id = "nvidia/canary-1b-v2"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, device_map="auto")

ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))

inputs = processor.apply_transcription_request(audio=ds[0]["audio"]["array"], source_language="en").to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=128)
print(processor.decode(generated_ids, skip_special_tokens=True)[0])

➡️ Translation

python

inputs = processor.apply_transcription_request(
    audio=ds[0]["audio"]["array"], source_language="en", target_language="de"
).to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=128)
print(processor.decode(generated_ids, skip_special_tokens=True)[0])

➡️ Batch inference

python

audios = [ds[0]["audio"]["array"], ds[1]["audio"]["array"]]
inputs = processor.apply_transcription_request(
    audio=audios, source_language="en", target_language=["en", "de"]
).to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=128)
for text in processor.decode(generated_ids, skip_special_tokens=True):
    print(text)

➡️ Training

Put the target transcript in the assistant turn and pass output_labels=True. Padding positions are masked automatically.

python

model.train()
transcription = "mister Quilter is the apostle of the middle classes, and we are glad to welcome his gospel."

conversation = [
    [
        {
            "role": "user",
            "content": [
                {"type": "audio", "audio": ds[0]["audio"]["array"]},
                {"type": "text", "source_language": "en", "target_language": "en", "punctuation": True},
            ],
        },
        {"role": "assistant", "content": transcription},
    ]
]

inputs = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
    processor_kwargs={"output_labels": True},
).to(model.device)

outputs = model(**inputs)
outputs.loss.backward()

For more details about usage, please refer to the Transformers' documentation.

Software Integration

Runtime Engine(s):

NeMo main branch (until it is released in NeMo 2.5)

Supported Hardware Microarchitecture Compatibility:

NVIDIA Ampere
NVIDIA Blackwell
NVIDIA Hopper

[Preferred/Supported] Operating System(s):

Linux

Hardware Specific Requirements: At least 6GB RAM for model to load.

Model Version

Current version: Canary-1b-v2. Previous versions can be accessed here.

Training and Evaluation Datasets

Training

The model was trained using the NeMo toolkit [4], following a 3-stage training procedure:

Initialized from a 4-language ASR model
Stage 1: Trained for 150,000 steps on X→En and English ASR tasks using 64 A100 GPUs
Stage 2: Trained for 115,000 additional steps on the full dataset (ASR, X→En, En→X)
Stage 3: Fine-tuned for 10,000 steps on a language-balanced high-quality subset of Granary and NeMo ASR Set 3.0

For all the stages of training, both languages and corpora are weighted using temperature sampling (τ = 0.5).

Training script: speech_to_text_aed.py

Tokenizer script: process_asr_text_tokenizer.py

Training Dataset

Canary-1b-v2 was trained on a massive multilingual speech recognition and translation dataset combining Nvidia's newly published Granary and in-house dataset NeMo ASR Set 3.0.

Granary Dataset [5] [6] with improved pseudo-labels and efficiently filtered versions of the following corpora:

YTC [7]
MOSEL [8]
YODAS [9]

Granary is now available on Hugging Face.

To read more about the pseudo-labeling technique and pipeline, please refer to the Granary Paper.

NeMo ASR Set 3.0 including human-labeled transcriptions from the following corpora:

Multilingual LibriSpeech (MLS)
Mozilla Common Voice (v7.0)
AMI (70 hrs)
Fleurs
LibriSpeech (960 hours)
Fisher Corpus
National Speech Corpus Part 1
VCTK
Europarl-ASR

Total training hours: 1.7M

ASR: 660,000 hrs
X→En: 360,000 hrs
En→X: 690,000 hrs
Non-speech: 36,000 hrs

All transcripts include punctuation and capitalization.

Data Collection Method by dataset

Hybrid: Automated, Human

Labeling Method by dataset

Hybrid: Synthetic, Human

Evaluation Dataset

Fleurs [10], MLS [11], CoVoST [12]
Hugging Face Open ASR Leaderboard [13]
Earnings-22 [14], This American Life [15] (long-form)
MUSAN [16]

Data Collection Method by dataset

Human

Labeling Method by dataset

Human

Benchmark Results

This section reports the evaluation results of the Canary-1b-v2 model across multiple tasks, including Automatic Speech Recognition (ASR), Speech Translation (AST), robustness to noise, and long-form transcription.

Automatic Speech Recognition (ASR)

WER ↓	Fleurs-25 Langs	CoVoST-13 Langs	MLS - 6 Langs
`Canary-1b-v2`	8.40%	8.85%	7.27%

Note: Presented WERs do not include Punctuation and Capitalization errors.

Hugging Face Open ASR Leaderboard

WER ↓	RTFx	Mean	AMI	GigaSpeech	LS Clean	LS Other	Earnings22	SPGISpech	Tedlium	Voxpopuli
`Canary-1b-v2`	749	7.15	16.01	10.82	2.18	3.56	11.79	2.28	4.29	6.25

More details on evaluation can be found at HuggingFace ASR Leaderboard

Speech Translation (AST)

X → English

	COMET ↑		BLEU ↑
	Fleurs-24 Langs	CoVoST-13 Langs	Fleurs-24 Langs	CoVoST-13 Langs
`Canary-1b-v2`	79.30	77.48	29.08	40.48

English → X

	COMET ↑		BLEU ↑
	Fleurs-24 Langs	CoVoST-5 Langs	Fleurs-24 Langs	CoVoST-5 Langs
`Canary-1b-v2`	84.56	80.29	29.4	32.33

Noise Robustness

Performance across different Signal-to-Noise Ratios (SNR) using MUSAN music and noise samples [16] on the LibriSpeech Clean test set. Metric: Word Error Rate (WER)

SNR (dB)	100	10	5	0	-5
`Canary-1b-v2`	2.18%	2.29%	2.80%	5.08%	19.38%

Hallucination Robustness

Number of characters per minute on MUSAN [16] 48 hrs eval set:

	# of character per minute ↓
`Canary-1b-v2`	134.7

Long-form Inference

Canary-1b-v2 achieves strong performance on long-form transcription by using dynamic chunking with 1-second overlap between chunks, allowing for efficient parallel processing. This dynamic chunking feature is automatically enabled when calling .transcribe() on a single audio file, or when using batch_size=1 with multiple audio files that are longer than 40 seconds.

Dataset	WER ↓
Earnings-22	13.78%
This American Life	9.87%

Note: Presented WERs do not include Punctuation and Capitalization errors.

Inference

Engine:

NVIDIA NeMo

Test Hardware:

NVIDIA A10
NVIDIA A100
NVIDIA A30
NVIDIA A5000
NVIDIA H100
NVIDIA L4
NVIDIA L40

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards here.

Please report security vulnerabilities or NVIDIA AI Concerns here.

Bias:

Field	Response
Participation considerations from adversely impacted groups protected classes in model design and testing	None
Measures taken to mitigate against unwanted bias	None

Explainability:

Field	Response
Intended Domain	Speech to Text Transcription and Translation
Model Type	Attention Encoder-Decoder
Intended Users	This model is intended for developers, researchers, academics, and industries building conversational based applications.
Output	Text
Describe how the model works	Speech input is encoded into embeddings and passed into conformer-based model and output a text response.
Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of	Not Applicable
Technical Limitations & Mitigation	Transcripts and translations may be not 100% accurate. Accuracy varies based on source and target language and characteristics of input audio (Domain, Use Case, Accent, Noise, Speech Type, Context of speech, etc.)
Verified to have met prescribed NVIDIA quality standards	Yes
Performance Metrics	Word Error Rate (Speech Transcription) / BLEU score (Speech Translation) / COMET score (Speech Translation)
Potential Known Risks	If a word is not trained in the language model and not presented in vocabulary, the word is not likely to be recognized. Not recommended for word-for-word/incomplete sentences as accuracy varies based on the context of input text
Licensing	GOVERNING TERMS: Use of this model is governed by the CC-BY-4.0 license.

Privacy:

Field	Response
Generatable or reverse engineerable personal data?	None
Personal data used to create this model?	None
Is there provenance for all datasets used in training?	Yes
Does data labeling (annotation, metadata) comply with privacy laws?	Yes
Is data compliant with data subject requests for data correction or removal, if such a request was made?	No, not possible with externally-sourced data.
Applicable Privacy Policy	https://www.nvidia.com/en-us/about-nvidia/privacy-policy/

Safety:

Field	Response
Model Application(s)	Speech to Text Transcription
Describe the life critical impact	None
Use Case Restrictions	Abide by CC-BY-4.0 License
Model and dataset restrictions	The Principle of least privilege (PoLP) is applied limiting access for dataset generation and model development. Restrictions enforce dataset access during training, and dataset license constraints adhered to.