README.md

---
license: cc-by-4.0
language:
- en
- de
- es
- fr
library_name: nemo
datasets:
- librispeech_asr
- fisher_corpus
- Switchboard-1
- WSJ-0
- WSJ-1
- National-Singapore-Corpus-Part-1
- National-Singapore-Corpus-Part-6
- vctk
- voxpopuli
- europarl
- multilingual_librispeech
- mozilla-foundation/common_voice_8_0
- MLCommons/peoples_speech
thumbnail: null
tags:
- automatic-speech-recognition
- automatic-speech-translation
- speech
- audio
- Transformer
- FastConformer
- Conformer
- pytorch
- NeMo
- hf-asr-leaderboard
widget:
- example_title: Librispeech sample 1
  src: https://cdn-media.huggingface.co/speech_samples/sample1.flac
- example_title: Librispeech sample 2
  src: https://cdn-media.huggingface.co/speech_samples/sample2.flac
model-index:
- name: canary-1b-flash
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: LibriSpeech (other)
      type: librispeech_asr
      config: other
      split: test
      args:
        language: en
    metrics:
    - name: Test WER
      type: wer
      value: 2.87
  - task:
      type: Automatic Speech Recognition
      name: automatic-speech-recognition
    dataset:
      name: SPGI Speech
      type: kensho/spgispeech
      config: test
      split: test
      args:
        language: en
    metrics:
    - name: Test WER
      type: wer
      value: 1.95
  - task:
      type: Automatic Speech Recognition
      name: automatic-speech-recognition
    dataset:
      name: Mozilla Common Voice 16.1
      type: mozilla-foundation/common_voice_16_1
      config: en
      split: test
      args:
        language: en
    metrics:
    - name: Test WER (En)
      type: wer
      value: 6.99
  - task:
      type: Automatic Speech Recognition
      name: automatic-speech-recognition
    dataset:
      name: Mozilla Common Voice 16.1
      type: mozilla-foundation/common_voice_16_1
      config: de
      split: test
      args:
        language: de
    metrics:
    - name: Test WER (De)
      type: wer
      value: 4.09
  - task:
      type: Automatic Speech Recognition
      name: automatic-speech-recognition
    dataset:
      name: Mozilla Common Voice 16.1
      type: mozilla-foundation/common_voice_16_1
      config: es
      split: test
      args:
        language: es
    metrics:
    - name: Test WER (ES)
      type: wer
      value: 3.62
  - task:
      type: Automatic Speech Recognition
      name: automatic-speech-recognition
    dataset:
      name: Mozilla Common Voice 16.1
      type: mozilla-foundation/common_voice_16_1
      config: fr
      split: test
      args:
        language: fr
    metrics:
    - name: Test WER (Fr)
      type: wer
      value: 6.15
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: en_us
      split: test
      args:
        language: en-de
    metrics:
    - name: Test BLEU (En->De)
      type: bleu
      value: 32.27
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: en_us
      split: test
      args:
        language: en-de
    metrics:
    - name: Test BLEU (En->Es)
      type: bleu
      value: 22.6
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: en_us
      split: test
      args:
        language: en-de
    metrics:
    - name: Test BLEU (En->Fr)
      type: bleu
      value: 41.22
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: de_de
      split: test
      args:
        language: de-en
    metrics:
    - name: Test BLEU (De->En)
      type: bleu
      value: 35.5
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: es_419
      split: test
      args:
        language: es-en
    metrics:
    - name: Test BLEU (Es->En)
      type: bleu
      value: 23.32
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: FLEURS
      type: google/fleurs
      config: fr_fr
      split: test
      args:
        language: fr-en
    metrics:
    - name: Test BLEU (Fr->En)
      type: bleu
      value: 33.42
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: COVOST
      type: covost2
      config: de_de
      split: test
      args:
        language: de-en
    metrics:
    - name: Test BLEU (De->En)
      type: bleu
      value: 39.33
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: COVOST
      type: covost2
      config: es_419
      split: test
      args:
        language: es-en
    metrics:
    - name: Test BLEU (Es->En)
      type: bleu
      value: 41.86
  - task:
      type: Automatic Speech Translation
      name: automatic-speech-translation
    dataset:
      name: COVOST
      type: covost2
      config: fr_fr
      split: test
      args:
        language: fr-en
    metrics:
    - name: Test BLEU (Fr->En)
      type: bleu
      value: 41.43
metrics:
- wer
- bleu
- comet
pipeline_tag: automatic-speech-recognition
---

# Canary 1B Flash

<style>
img {
 display: inline;
}
</style>

## Description:
NVIDIA NeMo Canary Flash [1] is a family of multilingual multi-tasking models based on Canary architecture [2] that achieve state-of-the-art performance on multiple speech benchmarks. With 883 million parameters and an inference speed of more than 1000 RTFx (on open-asr-leaderboard datasets), canary-1b-flash supports automatic speech-to-text recognition (ASR) in four languages (English, German, French, Spanish) and translation from English to German/French/Spanish and from German/French/Spanish to English with or without punctuation and capitalization (PnC). Additionally, canary-1b-flash offers an experimental feature for word-level and segment-level timestamps in English, German, French, and Spanish.
This model is released under the permissive CC-BY-4.0 license and is available for commercial use.


## Model Architecture:
Canary is an encoder-decoder model with FastConformer [3] Encoder and Transformer Decoder [4]. With audio features extracted from the encoder, task tokens such as \<target language\>, \<task\>, \<toggle timestamps\> and \<toggle PnC\> are fed into the Transformer Decoder to trigger the text generation process. Canary uses a concatenated tokenizer [5] from individual SentencePiece [6] tokenizers of each language, which makes it easy to scale up to more languages. The canary-1b-flash model has 32 encoder layers and 4 decoder layers, leading to a total of 883M parameters. For more details about the architecture, please refer to [1].

## NVIDIA NeMo

To train, fine-tune or transcribe with canary-1b-flash, you will need to install [NVIDIA NeMo](https://github.com/NVIDIA/NeMo).

## How to Use this Model

The model is available for use in the NeMo Framework [7], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Please refer to [our tutorial](https://github.com/NVIDIA/NeMo/blob/main/tutorials/asr/Canary_Multitask_Speech_Model.ipynb) for more details.

A few inference examples are listed below:

### Loading the Model

```python
from nemo.collections.asr.models import EncDecMultiTaskModel
# load model
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b-flash')
# update decode params
decode_cfg = canary_model.cfg.decoding
decode_cfg.beam.beam_size = 1
canary_model.change_decoding_strategy(decode_cfg)
```

## Input: 
**Input Type(s):** Audio <br>
**Input Format(s):** .wav or .flac files<br>
**Input Parameters(s):** 1D <br>
**Other Properties Related to Input:** 16000 Hz Mono-channel Audio, Pre-Processing Not Needed <br>

Input to canary-1b-flash can be either a list of paths to audio files or a jsonl manifest file.

If the input is a list of paths, canary-1b-flash assumes that the audio is English and transcribes it. I.e., canary-1b-flash default behavior is English ASR. 
```python
output = canary_model.transcribe(
    ['path1.wav', 'path2.wav'],
    batch_size=16,  # batch size to run the inference with
    pnc='yes',        # generate output with Punctuation and Capitalization
)

predicted_text_1 = output[0].text

```

canary-1b-flash can also generate word and segment level timestamps
```python
output = canary_model.transcribe(
  ['filepath.wav'],
  timestamps='yes',  # generate output with timestamps
)

predicted_text = output[0].text
word_level_timestamps = output[0].timestamp['word']
segment_level_timestamps = output[0].timestamp['segment']

```
For audio files longer than 10 seconds, we recommend using longform inference script (explained in next section) with `chunk_len_in_secs=10.0` to generate timestamps. 


To use canary-1b-flash for transcribing other supported languages or perform Speech-to-Text translation or provide word-level timestamps, specify the input as jsonl manifest file, where each line in the file is a dictionary containing the following fields: 

```yaml
# Example of a line in input_manifest.json
{
    "audio_filepath": "/path/to/audio.wav",  # path to the audio file
    "source_lang": "en",  # language of the audio input, set `source_lang`==`target_lang` for ASR, choices=['en','de','es','fr']
    "target_lang": "en",  # language of the text output, choices=['en','de','es','fr']
    "pnc": "yes",  # whether to have PnC output, choices=['yes', 'no']
    "timestamp": "yes", # whether to output word-level timestamps, choices=['yes', 'no']
}
```

and then use:
```python
output = canary_model.transcribe(
    "<path to input manifest file>",
    batch_size=16,  # batch size to run the inference with
)
```

### Longform inference with Canary-1B-flash:
Canary models are designed to handle input audio smaller than 40 seconds. In order to handle longer audios, NeMo includes [speech_to_text_aed_chunked_infer.py](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/asr_chunked_inference/aed/speech_to_text_aed_chunked_infer.py) script that handles chunking, performs inference on the chunked files, and stitches the transcripts.

The script will perform inference on all `.wav` files in `audio_dir`. Alternatively you can also pass a path to a manifest file as shown above. The decoded output will be saved at `output_json_path`.

```
python scripts/speech_to_text_aed_chunked_infer.py \
    pretrained_name="nvidia/canary-1b-flash" \
    audio_dir=$audio_dir \
    output_filename=$output_json_path \
    chunk_len_in_secs=40.0 \
    batch_size=1 \
    decoding.beam.beam_size=1 \
    timestamps=False
```

**Note** that for longform inference with timestamps, it is recommended to use `chunk_len_in_secs` of 10 seconds.


## Output:
**Output Type(s):** Text <br>
**Output Format:** Text output as a string (w/ timestamps) depending on the task chosen for decoding <br> 
**Output Parameters:** 1-Dimensional text string <br>
**Other Properties Related to Output:** May Need Inverse Text Normalization; Does Not Handle Special Characters <br>


## Software Integration:
**Runtime Engine(s):** 
* NeMo - main <br>

**Supported Hardware Microarchitecture Compatibility:** <br>
* [NVIDIA Ampere] <br>
* [NVIDIA Blackwell] <br>
* [NVIDIA Jetson]  <br>
* [NVIDIA Hopper] <br>
* [NVIDIA Lovelace] <br>
* [NVIDIA Pascal] <br>
* [NVIDIA Turing] <br>
* [NVIDIA Volta] <br>

**[Preferred/Supported] Operating System(s):** <br>
* [Linux] <br>
* [Linux 4 Tegra] <br>
* [Windows] <br>

## Model Version(s): 
canary-1b-flash <br>


# Training and Evaluation Datasets: 

## Training Dataset:

The canary-1b-flash model is trained on a total of 85K hrs of speech data. It consists of 31K hrs of public data, 20K hrs collected by [Suno](https://suno.ai/), and 34K hrs of in-house data. 
The datasets below include conversations, videos from the web and audiobook recordings.

**Data Collection Method:**
* Human <br>

**Labeling Method:**
* Hybrid: Human, Automated <br>

The constituents of public data are as follows. 

#### English (25.5k hours)
- Librispeech 960 hours
- Fisher Corpus
- Switchboard-1 Dataset
- WSJ-0 and WSJ-1
- National Speech Corpus (Part 1, Part 6)
- VCTK
- VoxPopuli (EN)
- Europarl-ASR (EN)
- Multilingual Librispeech (MLS EN) - 2,000 hour subset
- Mozilla Common Voice (v7.0)
- People's Speech - 12,000 hour subset
- Mozilla Common Voice (v11.0)  - 1,474 hour subset

#### German (2.5k hours)
- Mozilla Common Voice (v12.0)  - 800 hour subset
- Multilingual Librispeech (MLS DE) - 1,500 hour subset
- VoxPopuli (DE) - 200 hr subset

#### Spanish (1.4k hours)
- Mozilla Common Voice (v12.0)  - 395 hour subset
- Multilingual Librispeech (MLS ES) - 780 hour subset
- VoxPopuli (ES) - 108 hour subset
- Fisher  - 141 hour subset

#### French (1.8k hours)
- Mozilla Common Voice (v12.0)  - 708 hour subset
- Multilingual Librispeech (MLS FR) - 926 hour subset
- VoxPopuli (FR) - 165 hour subset


## Evaluation Dataset:

**Data Collection Method:** <br>
* Human <br>

**Labeling Method:** <br>
* Human <br>

Automatic Speech Recognition: 
* [HuggingFace OpenASR Leaderboard evaluation sets](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard)
* [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech)
* [MCV] (https://commonvoice.mozilla.org/en/datasets)

Automatic Speech Translation:
* [FLEURS](https://huggingface.co/datasets/google/fleurs)
* [COVOST-v2](https://github.com/facebookresearch/covost)
* [mExpresso](https://huggingface.co/facebook/seamless-expressive#mexpresso-multilingual-expresso)

Timestamp Prediction:
* [Librispeech](https://www.openslr.org/12)

Hallucination Robustness:
* [MUSAN](https://www.openslr.org/17/) 48 hrs eval set

Noise Robustness:
* [Librispeech](https://www.openslr.org/12)

Model Fairness:
* [Casual Conversations Dataset](https://arxiv.org/abs/2104.02821)

## Training

Canary-1B-Flash is trained using the NVIDIA NeMo Framework [7] for a total of 200K steps with 2D bucketing [1] and optimal batch sizes set using OOMptimizer [8].The model is trained on 128 NVIDIA A100 80GB GPUs. 
The model can be trained using this [example script](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/speech_multitask/speech_to_text_aed.py) and [base config](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/conf/speech_multitask/fast-conformer_aed.yaml).

The tokenizers for these models were built using the text transcripts of the train set with this [script](https://github.com/NVIDIA/NeMo/blob/main/scripts/tokenizers/process_asr_text_tokenizer.py).

## Inference:
**Engine:** NVIDIA NeMo <br>
**Test Hardware :** <br>
* A6000 <br>
* A100 <br>
* V100 <br>

## Performance

For ASR and AST experiments, predictions were generated using greedy decoding. Note that utterances shorter than 1 second are symmetrically zero-padded upto 1 second during evaluation.  

### English ASR Performance (w/o PnC) 

The ASR performance is measured with word error rate (WER), and we process the groundtruth and predicted text with [whisper-normalizer](https://pypi.org/project/whisper-normalizer/).

WER on [HuggingFace OpenASR leaderboard](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard):

| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   |
|:---------:|:-----------:|:------:|:------:|:------:|:------:|:------:|:------:|:------:|:------:|:------:|
| nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |

#### Inference speed on different systems
We profiled inference speed on the OpenASR benchmark (batch_size=128) using the [real-time factor](https://github.com/NVIDIA/DeepLearningExamples/blob/master/Kaldi/SpeechRecognition/README.md#metrics) (RTFx) to quantify throughput. 

| **Version** | **Model**     | **System**   | **RTFx**   |
|:-----------:|:-------------:|:------------:|:----------:|
| nemo-main | canary-1b-flash | NVIDIA A100 | 1045.75 |
| nemo-main | canary-1b-flash | NVIDIA H100 | 1669.07 |



### Multilingual ASR Performance

WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set:

| **Version** | **Model**  | **De**   | **Es**   | **Fr**   |
|:---------:|:-----------:|:------:|:------:|:------:|
| nemo-main   | canary-1b-flash | 4.36 | 2.69 | 4.47 |

WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set:
| **Version** | **Model**  |  **En**   | **De**   | **Es**   | **Fr**   |
|:---------:|:-----------:|:------:|:------:|:------:|:------:|
| nemo-main   | canary-1b-flash | 6.99 | 4.09 | 3.62 | 6.15 |


More details on evaluation can be found at [HuggingFace ASR Leaderboard](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard)

### AST Performance

We evaluate AST performance with [BLEU score](https://lightning.ai/docs/torchmetrics/stable/text/sacre_bleu_score.html) and [COMET score](https://aclanthology.org/2020.emnlp-main.213/), and use native annotations with punctuation and capitalization in the datasets.

[FLEURS](https://huggingface.co/datasets/google/fleurs) test set:

BLEU score:
| **Version** | **Model** | **En->De** | **En->Es** | **En->Fr** | **De->En** | **Es->En** | **Fr->En** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash | 	32.27   |  22.6   |   41.22    |   35.5   |   23.32    |   33.42    |

COMET score:
| **Version** | **Model** | **En->De** | **En->Es** | **En->Fr** | **De->En** | **Es->En** | **Fr->En** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash | 	0.8114   |  0.8118   |   0.8165    |   0.8546   |   0.8228   |   0.8475    |

[COVOST-v2](https://github.com/facebookresearch/covost) test set:

BLEU score:
| **Version** | **Model** | **De->En** | **Es->En** | **Fr->En** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |   39.33   |    41.86    |   41.43    |

COMET score:
| **Version** | **Model** | **De->En** | **Es->En** | **Fr->En** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |   0.8553   |    0.8585    |   0.8511    |

[mExpresso](https://huggingface.co/facebook/seamless-expressive#mexpresso-multilingual-expresso) test set:

BLEU score:
| **Version** | **Model** | **En->De** | **En->Es** | **En->Fr** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |   22.91    |   35.69   |   27.85   |

COMET score:
| **Version** | **Model** | **En->De** | **En->Es** | **En->Fr** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |   0.7889    |   0.8211   |   0.7910   |

### Timestamp Prediction
F1-score on [Librispeech Test sets](https://www.openslr.org/12) at collar value of 200ms
| **Version** | **Model** | **test-clean** | **test-other** |
|:-----------:|:---------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |   95.5    |   93.5   |


### Hallucination Robustness
Number of characters per minute on [MUSAN](https://www.openslr.org/17) 48 hrs eval set
| **Version** | **Model** | **# of character per minute** |
|:-----------:|:---------:|:----------:|
| nemo-main       | canary-1b-flash |   60.92   |

### Noise Robustness
WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise

| **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** |
|:-----------:|:---------:|:----------:|:----------:|:----------:|:----------:|
| nemo-main       | canary-1b-flash |    2.34   |   3.69   |   8.84   |    29.71  |

## Model Fairness Evaluation

As outlined in the paper "Towards Measuring Fairness in AI: the Casual Conversations Dataset" [9], we assessed the canary-1b-flash model for fairness. The model was evaluated on the CausalConversations-v1 dataset, and the results are reported as follows:

### Gender Bias:

| Gender | Male | Female | N/A | Other |
| :--- | :--- | :--- | :--- | :--- |
| Num utterances | 19325 | 24532 | 926 | 33 |
| % WER | 14.66 | 12.44 | 17.17 | 27.56 |

### Age Bias:

| Age Group | (18-30) | (31-45) | (46-85) | (1-100) |
| :--- | :--- | :--- | :--- | :--- |
| Num utterances | 15956 | 14585 | 13349 | 43890 |
| % WER | 13.18 | 13.45 | 13.64 | 13.41 |

(Error rates for fairness evaluation are determined by normalizing both the reference and predicted text, similar to the methods used in the evaluations found at https://github.com/huggingface/open_asr_leaderboard.)

## License/Terms of Use: 
canary-1b-flash is released under the CC-BY-4.0 license. By using this model, you are agreeing to the [terms and conditions](https://choosealicense.com/licenses/cc-by-4.0/) of the license. <br>

## References:
[1] [Training and Inference Efficiency of Encoder-Decoder Speech Models](https://arxiv.org/abs/2503.05931)

[2] [Less is More: Accurate Speech Recognition & Translation without Web-Scale Data](https://www.isca-archive.org/interspeech_2024/puvvada24_interspeech.pdf)

[3] [Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition](https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10389701)

[4] [Attention is All You Need](https://arxiv.org/abs/1706.03762)

[5] [Unified Model for Code-Switching Speech Recognition and Language Identification Based on Concatenated Tokenizer](https://aclanthology.org/2023.calcs-1.7.pdf)

[6] [Google Sentencepiece Tokenizer](https://github.com/google/sentencepiece)

[7] [NVIDIA NeMo Framework](https://github.com/NVIDIA/NeMo)

[8] [EMMeTT: Efficient Multimodal Machine Translation Training](https://arxiv.org/abs/2409.13523)

[9] [Towards Measuring Fairness in AI: the Casual Conversations Dataset](https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9634168)


## 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 internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.  
Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).