nvidia/canary-1b

Canary 1B

| |

NVIDIA NeMo Canary is a family of multi-lingual multi-tasking models that achieves state-of-the art performance on multiple benchmarks. With 1 billion parameters, Canary-1B supports automatic speech-to-text recognition (ASR) in 4 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).

Model Architecture

Canary is an encoder-decoder model with FastConformer [1] encoder and Transformer Decoder [2]. With audio features extracted from the encoder, task tokens such as <source language>, <target language>, <task> and <toggle PnC> are fed into the Transformer Decoder to trigger the text generation process. Canary uses a concatenated tokenizer [5] from individual SentencePiece [3] tokenizers of each language, which makes it easy to scale up to more languages. The Canay-1B model has 24 encoder layers and 24 layers of decoder layers in total.

NVIDIA NeMo

To train, fine-tune or Transcribe with Canary, you will need to install NVIDIA NeMo. We recommend you install it after you've installed Cython and latest PyTorch version.

pip install git+https://github.com/NVIDIA/NeMo.git@r1.23.0#egg=nemo_toolkit[asr]

How to Use this Model

The model is available for use in the NeMo toolkit [4], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Loading the Model

from nemo.collections.asr.models import EncDecMultiTaskModel

# load model
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')

# update dcode params
decode_cfg = canary_model.cfg.decoding
decode_cfg.beam.beam_size = 1
canary_model.change_decoding_strategy(decode_cfg)

Input Format

Input to Canary can be either a list of paths to audio files or a jsonl manifest file.

If the input is a list of paths, Canary assumes that the audio is English and Transcribes it. I.e., Canary default behaviour is English ASR.

predicted_text = canary_model.transcribe(
    paths2audio_files=['path1.wav', 'path2.wav'],
    batch_size=16,  # batch size to run the inference with
)

To use Canary for transcribing other supported languages or perform Speech-to-Text translation, specify the input as jsonl manifest file, where each line in the file is a dictionary containing the following fields:

# Example of a line in input_manifest.json
{
    "audio_filepath": "/path/to/audio.wav",  # path to the audio file
    "duration": 1000,  # duration of the audio, can be set to `None` if using NeMo main branch
    "taskname": "asr",  # use "s2t_translation" for speech-to-text translation with r1.23, or "ast" if using the NeMo main branch
    "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']
    "answer": "na", 
}

and then use:

predicted_text = canary_model.transcribe(
    "<path to input manifest file>",
    batch_size=16,  # batch size to run the inference with
)

Automatic Speech-to-text Recognition (ASR)

An example manifest for transcribing English audios can be:

# Example of a line in input_manifest.json
{
    "audio_filepath": "/path/to/audio.wav",  # path to the audio file
    "duration": 1000,  # duration of the audio, can be set to `None` if using NeMo main branch
    "taskname": "asr",  
    "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']
    "answer": "na", 
}

Automatic Speech-to-text Translation (AST)

An example manifest for transcribing English audios into German text can be:

# Example of a line in input_manifest.json
{
    "audio_filepath": "/path/to/audio.wav",  # path to the audio file
    "duration": 1000,  # duration of the audio, can be set to `None` if using NeMo main branch
    "taskname": "s2t_translation", # r1.23 only recognizes "s2t_translation", but "ast" is supported if using the NeMo main branch
    "source_lang": "en", # language of the audio input, choices=['en','de','es','fr']
    "target_lang": "de", # language of the text output, choices=['en','de','es','fr']
    "pnc": "yes",  # whether to have PnC output, choices=['yes', 'no']
    "answer": "na" 
}

Alternatively, one can use transcribe_speech.py script to do the same.

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py 
 pretrained_name="nvidia/canary-1b" 
 audio_dir="<path to audio_directory>" # transcribes all the wav files in audio_directory

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py 
 pretrained_name="nvidia/canary-1b" 
 dataset_manifest="<path to manifest file>" 

Input

This model accepts single channel (mono) audio sampled at 16000 Hz, along with the task/languages/PnC tags as input.

Output

The model outputs the transcribed/translated text corresponding to the input audio, in the specified target language and with or without punctuation and capitalization.

Training

Canary-1B is trained using the NVIDIA NeMo toolkit [4] for 150k steps with dynamic bucketing and a batch duration of 360s per GPU on 128 NVIDIA A100 80GB GPUs. The model can be trained using this example script and base config.

The tokenizers for these models were built using the text transcripts of the train set with this script.

Datasets

The Canary-1B 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, and 34k hrs of in-house data.

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

Performance

In both ASR and AST experiments, predictions were generated using beam search with width 5 and length penalty 1.0.

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.

WER on MCV-16.1 test set:

Version	Model	En	De	Es	Fr
1.23.0	canary-1b	7.97	4.61	3.99	6.53

WER on MLS test set:

Version	Model	En	De	Es	Fr
1.23.0	canary-1b	3.06	4.19	3.15	4.12

More details on evaluation can be found at HuggingFace ASR Leaderboard

AST Performance

We evaluate AST performance with BLEU score, and use native annotations with punctuation and capitalization in the datasets.

BLEU score on FLEURS test set:

Version	Model	En->De	En->Es	En->Fr	De->En	Es->En	Fr->En
1.23.0	canary-1b	32.15	22.66	40.76	33.98	21.80	30.95

BLEU score on COVOST-v2 test set:

Version	Model	De->En	Es->En	Fr->En
1.23.0	canary-1b	37.67	40.7	40.42

BLEU score on mExpresso test set:

Version	Model	En->De	En->Es	En->Fr
1.23.0	canary-1b	23.84	35.74	28.29

Model Fairness Evaluation

As outlined in the paper "Towards Measuring Fairness in AI: the Casual Conversations Dataset", we assessed the canary-1.1b 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	14.64	12.92	17.88	126.92

Age Bias:

Age Group	(18-30)	(31-45)	(46-85)	(1-100)
Num utterances	15956	14585	13349	43890
% WER	14.64	13.07	13.47	13.76

(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.)

NVIDIA Riva: Deployment

NVIDIA Riva, is an accelerated speech AI SDK deployable on-prem, in all clouds, multi-cloud, hybrid, on edge, and embedded. Additionally, Riva provides:

• World-class out-of-the-box accuracy for the most common languages with model checkpoints trained on proprietary data with hundreds of thousands of GPU-compute hours
• Best in class accuracy with run-time word boosting (e.g., brand and product names) and customization of acoustic model, language model, and inverse text normalization
• Streaming speech recognition, Kubernetes compatible scaling, and enterprise-grade support

Although this model isn’t supported yet by Riva, the list of supported models is here.
Check out Riva live demo.

References

[1] Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

[2] Attention is all you need

[3] Google Sentencepiece Tokenizer

[4] NVIDIA NeMo Toolkit

[5] Unified Model for Code-Switching Speech Recognition and Language Identification Based on Concatenated Tokenizer

Licence

License to use this model is covered by the CC-BY-NC-4.0. By downloading the public and release version of the model, you accept the terms and conditions of the CC-BY-NC-4.0 license.