Model Overview

Citrinet-512 model which has been trained on the ASR Set dataset with over 7000 hours of english speech.

It utilizes a Google SentencePiece [1] tokenizer with vocabulary size 1024, and transcribes text in lower case english alphabet along with spaces, apostrophes and a few other characters.

Model Architecture

Citrinet is a deep residual convolutional neural network architecture that is optimized for Automatic Speech Recognition tasks. There are many variants of the Citrinet family of models, which are further discussed in the paper [2].

Training

These models were trained on a composite dataset comprising of several thousand hours of speech, compiled from various publicly available sources. The NeMo toolkit [3] was used for training this model over several hundred epochs on multiple GPUs.

Datasets

While training this model, we used the following datasets:

• Librispeech 960 hours of English speech
• Fisher Corpus
• Switchboard-1 Dataset
• WSJ-0 and WSJ-1
• National Speech Corpus - 1
• Mozilla Common Voice

Tokenizer Construction

The tokenizer for this model was built using text corpus provided with the train dataset.

We build a Google Sentencepiece Tokenizer [1] with the following script :

python [NEMO_GIT_FOLDER]/scripts/tokenizers/process_asr_text_tokenizer.py \
  --manifest="train_manifest.json" \
  --data_root="<OUTPUT DIRECTORY FOR TOKENIZER>" \
  --vocab_size=1024 \
  --tokenizer="spe" \
  --spe_type="unigram" \
  --spe_character_coverage=1.0 \
  --no_lower_case \
  --log

Performance

The performance of Automatic Speech Recognition models is measuring using Word Error Rate. Since this dataset is trained on multiple domains and a much larger corpus, it will generally perform better at transcribing audio in general.

The model obtains the following scores on the following evaluation datasets -

• 3.7 % on Librispeech dev-clean
• 8.9 % on Librispeech dev-other
• 3.7 % on Librispeech test-clean
• 8.9 % on Librispeech test-other
• 3.2 % on WSJ Eval 92
• 4.3 % on WSJ Dev 93
• 7.5 % on NSC Part 1

Note that these scores on Librispeech are not particularly indicative of the quality of transcriptions that models trained on ASR Set will achieve, but they are a useful proxy.

How to Use this Model

The model is available for use in the NeMo toolkit [3], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Automatically load the model from NGC

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained(model_name="stt_en_citrinet_512")

Transcribing text with this model

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py \
  pretrained_name="stt_en_citrinet_512" \
  audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"

Input

This model accepts 16000 KHz Mono-channel Audio (wav files) as input.

Output

This model provides transcribed speech as a string for a given audio sample.

Limitations

Since this model was trained on publically available speech datasets, the performance of this model might degrade for speech which includes technical terms, or vernacular that the model has not been trained on. The model might also perform worse for accented speech.

References

[1] Google Sentencepiece Tokenizer

[2] Citrinet: Closing the Gap between Non-Autoregressive and Autoregressive End-to-End Models for Automatic Speech Recognition

[3] NVIDIA NeMo Toolkit

Licence

License to use this model is covered by the NGC TERMS OF USE unless another License/Terms Of Use/EULA is clearly specified. By downloading the public and release version of the model, you accept the terms and conditions of the NGC TERMS OF USE.