README.md

metadata

language:
  - es
tags:
  - audio
  - automatic-speech-recognition
datasets:
  - multilingual_librispeech
metrics:
  - eval_wer: 0.073
model-index:
  - name: wav2vec2-spanish-multilibrispeech
    results:
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          type: multilingual_librispeech
          name: multilingual_librispeech es
          args: es
        metrics:
          - type: wer
            value: 0.073
            name: eval_wer
          - type: loss
            value: 0.086
            name: eval_loss

This is a model for automatic speech recognition in spanish, by using the Spanish portion of multilingual_librispeech and the pre-trained wav2vec2 multilingual from Facebook

For training the model, we used the same parameters as they recommend in the paper. We trained for a total of 15 epochs, obtaining a final wer of 0.073.

An example of how to use this model:

from transformers import Wav2Vec2Tokenizer, AutoModelForCTC

tokenizer = Wav2Vec2Tokenizer.from_pretrained(
            "IIC/wav2vec2-spanish-multilibrispeech"
        )

model = AutoModelForCTC.from_pretrained(
            "IIC/wav2vec2-spanish-multilibrispeech"
        )

Contributions

Thanks to @avacaondata, @alborotis, @albarji, @Dabs, @GuillemGSubies for adding this model.