--- license: cc-by-nc-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 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: 2.89 - 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: 4.79 - 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: 7.97 - 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: 4.61 - 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: 3.99 - task: type: Automatic Speech Recognition name: automatic-speech-recognition dataset: name: Mozilla Common Voice 16.1 type: mozilla-foundation/common_voice_16_1 config: fr split: test args: language: fr metrics: - name: Test WER (Fr) type: wer value: 6.53 - 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: 22.66 - 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: 41.11 - 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: 40.76 - 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: 32.64 - 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: 32.15 - 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: 23.57 - 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: 37.67 - 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: 40.7 - 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: 40.42 metrics: - wer - bleu pipeline_tag: automatic-speech-recognition --- # Canary 1B [![Model architecture](https://img.shields.io/badge/Model_Arch-FastConformer--Transformer-lightgrey#model-badge)](#model-architecture) | [![Model size](https://img.shields.io/badge/Params-1B-lightgrey#model-badge)](#model-architecture) | [![Language](https://img.shields.io/badge/Language-en-lightgrey#model-badge)](#datasets) | [![Language](https://img.shields.io/badge/Language-de-lightgrey#model-badge)](#datasets) | [![Language](https://img.shields.io/badge/Language-es-lightgrey#model-badge)](#datasets) | [![Language](https://img.shields.io/badge/Language-fr-lightgrey#model-badge)](#datasets) 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 ``, ``, `` and `` are fed into the Transformer Decoder to trigger the text generation process. Canary uses a concatenated tokenizer 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 play with the model you will need to install [NVIDIA NeMo](https://github.com/NVIDIA/NeMo). We recommend you install it after you've installed latest PyTorch version. ``` pip install nemo_toolkit['all'] ``` ## 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 ```python 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 = 5 # default is greedy with beam_size=1 canary_model.change_decoding_strategy(decode_cfg) ``` ### Input Format The input to the model can be a directory containing audio files, in which case the model will perform ASR on English and produces text with punctuation and capitalization: ```python predicted_text = canary_model.trancribe( audio_dir="", batch_size=16, # batch size to run the inference with ) ``` or use: ```bash python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py pretrained_name="nvidia/canary-1b" audio_dir="" ``` Another recommended option is to use a json manifest as input, 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 "duration": 10000.0, # duration of the audio "taskname": "asr", # use "s2t_translation" for AST "source_lang": "en", # Set `source_lang`=`target_lang` for ASR, choices=['en','de','es','fr'] "target_lang": "de", # choices=['en','de','es','fr'] "pnc": yes, # whether to have PnC output, choices=['yes', 'no'] } ``` and then use: ```python predicted_text = canary_model.trancribe( paths2audio_files="", batch_size=16, # batch size to run the inference with ) ``` or use: ```bash python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py pretrained_name="nvidia/canary-1b" dataset_manifest="" ``` ### Automatic Speech-to-text Recognition (ASR) An example manifest for transcribing English audios can be: ```yaml # Example of a line in input_manifest.json { "audio_filepath": "/path/to/audio.wav", # path to the audio file "duration": 10000.0, # duration of the audio "taskname": "asr", "source_lang": "en", "target_lang": "en", "pnc": yes, # whether to have PnC output, choices=['yes', 'no'] } ``` ### Automatic Speech-to-text Translation (AST) An example manifest for transcribing English audios into German text can be: ```yaml # Example of a line in input_manifest.json { "audio_filepath": "/path/to/audio.wav", # path to the audio file "duration": 10000.0, # duration of the audio "taskname": "s2t_translation", "source_lang": "en", "target_lang": "de", "pnc": yes, # whether to have PnC output, choices=['yes', 'no'] } ``` ### 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 in 24 hrs. 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](https://github.com/NVIDIA/NeMo/blob/main/scripts/tokenizers/process_asr_text_tokenizer.py). ### Datasets The Canary-1B model is trained on 70K hours of speech audio with transcriptions in their original languages for ASR, and machine-generated translations for each supported language for speech translation. The training data contains 43K hours of English speech collected and prepared by NVIDIA NeMo and [Suno](https://suno.ai/) teams, and an inhouse subset with 27K hours of English/German/Spanish/French speech. ## 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) on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test sets on four languages, and we process the groundtruth and predicted text with [whisper-normalizer](https://pypi.org/project/whisper-normalizer/). | **Version** | **Model** | **En** | **De** | **Es** | **Fr** | |:---------:|:-----------:|:------:|:------:|:------:|:------:| | 1.23.0 | canary-1b | 7.97 | 4.61 | 3.99 | 6.53 | 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 on the [FLEURS](https://huggingface.co/datasets/google/fleurs) test sets on four languages and use their native annotations with punctuation and capitalization. | **Version** | **Model** | **En->De** | **En->Es** | **En->Fr** | **De->En** | **Es->En** | **Fr->En** | |:-----------:|:---------:|:----------:|:----------:|:----------:|:----------:|:----------:|:----------:| | 1.23.0 | canary-1b | 22.66 | 41.11 | 40.76 | 32.64 | 32.15 | 23.57 | ## NVIDIA Riva: Deployment [NVIDIA Riva](https://developer.nvidia.com/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](https://huggingface.co/models?other=Riva) is here. Check out [Riva live demo](https://developer.nvidia.com/riva#demos). ## References [1] [Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition](https://arxiv.org/abs/2305.05084) [2] [Attention is all you need](https://arxiv.org/abs/1706.03762) [3] [Google Sentencepiece Tokenizer](https://github.com/google/sentencepiece) [4] [NVIDIA NeMo Toolkit](https://github.com/NVIDIA/NeMo) ## Licence License to use this model is covered by the [CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/deed.en#:~:text=NonCommercial%20%E2%80%94%20You%20may%20not%20use,doing%20anything%20the%20license%20permits.). By downloading the public and release version of the model, you accept the terms and conditions of the CC-BY-NC-4.0 license.