STT Es Conformer Transducer Large

Model Overview

Conformer-Transducer model which was pre-trained in a self supervised manner [11] and then finetuned on Spanish labelled speech data. The model was pretrained on a subset of the unlabeled Voxpopuli [4] dataset. Only the English, German, Spanish and French unlabelled subsets were used, which amounted to 100,000 hours. Then the model was finetuned using labelled Spanish portions of the Mozilla CommonVoice (MCV7.0) [2], Multilingual LibriSpeech (MLS) [3], and Voxpopuli [4] training sets, as well as a subset of the Spanish Fisher dataset [5, 6]. This amounted to 1,340 hours of Spanish training data.

The model utilizes a Google SentencePiece [7] tokenizer with vocabulary size 128, and transcribes text in lower case Spanish alphabet characters.

Model Architecture

Conformer-Transducer model is an autoregressive variant of Conformer model [1] for Automatic Speech Recognition which uses Transducer loss/decoding. You may find more info on the detail of this model here: Conformer-Transducer Model.

Training

The NeMo toolkit [8] was used for training this model over several hundred epochs on multiple GPUs.

Datasets

While pre-training this model (in a self-supervised manner), we used the following datasets:

• Voxpopuli unlabelled speech data for English, German, Spanish and French - 100,000 hours

While finetuning this model, we used the following datasets:

• Mozilla Common Voice 7.0 (Spanish) - 289 hours of training data after data cleaning
• Multilingual LibriSpeech (Spanish) - 801 hours of training data after data cleaning
• Voxpopuli transcribed subset (Spanish) - 110 hours of training data after data cleaning
• Fisher dataset (Spanish) - 140 hours of training data after data cleaning

Tokenizer Construction

The tokenizer for this model was built using the text that was in the training set.

We build a Google Sentencepiece Tokenizer with the following script:

python [NEMO_GIT_FOLDER]/scripts/process_asr_text_tokenizer.py \
  --manifest="train_manifest.json" \
  --data_root="<OUTPUT DIRECTORY FOR TOKENIZER>" \
  --vocab_size=128 \
  --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.

The model obtains the following greedy scores on the following evaluation datasets:

• 4.6 % WER on MCV7.0 dev set
• 2.7 % WER on MLS dev set
• 4.7 % WER on Voxpopuli dev set
• 14.7 % WER on Fisher dev set (manually created)

• 5.2 % WER on MCV7.0 test set
• 3.2 % WER on MLS test set
• 6.0 % WER on Voxpopuli test set
• 14.8 % WER on Fisher test set (manually created)

• 17.2 % WER on Spanish CALLHOME train set [9,10] (model was not trained on this data)
• 18.0 % WER on Spanish CALLHOME devtest set [9,10] (model was not trained on this data)
• 19.4 % WER on Spanish CALLHOME evltest set [9,10] (model was not trained on this data)

How to Use this Model

The model is available for use in the NeMo toolkit [5], 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_es_conformer_transducer_large")

Transcribing text with this model

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py \
  pretrained_name="stt_es_conformer_transducer_large" \
  audio_dir=""

Input

This model accepts 16000 Hz 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] Conformer: Convolution-augmented Transformer for Speech Recognition

[2] Mozilla CommonVoice (MCV7.0)

[3] Multilingual LibriSpeech (MLS)

[4] VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation

[5] Fisher Spanish - Transcripts

[6] Fisher Spanish Speech

[7] Google Sentencepiece Tokenizer

[8] NVIDIA NeMo Toolkit

[9] CALLHOME Spanish Transcripts

[10] CALLHOME Spanish Speech

[11] Self-Supervised pre-training for ASR in NeMo

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.