metadata

language: br
datasets:
  - common_voice
tags:
  - audio
  - automatic-speech-recognition
  - speech
  - xlsr-fine-tuning-week
license: apache-2.0
model-index:
  - name: XLSR Wav2Vec2 Breton by Marxav
    results:
      - task:
          name: Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice br
          type: common_voice
          args: br
        metrics:
          - name: Test WER
            type: wer
            value: 44.34

Wav2Vec2-Large-XLSR-53-Breton

The model can be used directly (without a language model) as follows:

TODO

The model can be evaluated as follows on the {language} test data of Common Voice.

import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re

test_dataset = load_dataset("common_voice", "br", split="test")
wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained('Marxav/wav2vec2-large-xlsr-53-breton')
model = Wav2Vec2ForCTC.from_pretrained('Marxav/wav2vec2-large-xlsr-53-breton')
model.to("cuda")


chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\'\(\)\/\«\»\½\…]'

def remove_special_characters(batch):
     sentence = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower() + " "
     sentence = re.sub("ʼ","'", sentence)
     sentence = re.sub("’","'", sentence)
     batch["sentence"] = sentence
     return batch

resampler = torchaudio.transforms.Resample(48_000, 16_000)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
  batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
  speech_array, sampling_rate = torchaudio.load(batch["path"])
  batch["speech"] = resampler(speech_array).squeeze().numpy()
  return batch

test_dataset = test_dataset.map(remove_special_characters)

test_dataset = test_dataset.map(speech_file_to_array_fn)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def evaluate(batch):
  inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)

  with torch.no_grad():
    logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits

    pred_ids = torch.argmax(logits, dim=-1)
  batch["pred_strings"] = processor.batch_decode(pred_ids)
  return batch

result = test_dataset.map(evaluate, batched=True, batch_size=8)

print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))

Test Result: 44.34%

Training

The Common Voice train, validation datasets were used for training.