Fine-tuned facebook/wav2vec2-large-xlsr-53 on the Irish Common Voice dataset. When using this model, make sure that your speech input is sampled at 16kHz.

Usage

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

python

import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "ga-IE", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("jimregan/wav2vec2-large-xlsr-irish-basic")
model = Wav2Vec2ForCTC.from_pretrained("jimregan/wav2vec2-large-xlsr-irish-basic")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the audio files as arrays
def speech_file_to_array_fn(batch):
	speech_array, sampling_rate = torchaudio.load(batch["path"])
	batch["speech"] = resampler(speech_array).squeeze().numpy()
	return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
	logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])

Evaluation

The model can be evaluated as follows on the Irish test data of Common Voice.

python

import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "ga-IE", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("jimregan/wav2vec2-large-xlsr-irish-basic")
model = Wav2Vec2ForCTC.from_pretrained("jimregan/wav2vec2-large-xlsr-irish-basic") 
model.to("cuda")
# So, tolower() for Irish is a bit complicated: tAthar -> t-athair
# toupper() is non-deterministic :)
def is_upper_vowel(letter):
    if letter in ['A', 'E', 'I', 'O', 'U', 'Á', 'É', 'Í', 'Ó', 'Ú']:
        return True
    else:
        return False
def irish_lower(word):
    if len(word) > 1 and word[0] in ['n', 't'] and is_upper_vowel(word[1]):
        return word[0] + '-' + word[1:].lower()
    else:
        return word.lower()
def irish_lower_sentence(sentence):
    return " ".join([irish_lower(w) for w in sentence.split(" ")])
chars_to_ignore_regex = '[,\?\.\!\;\:\"\“\%\‘\”\(\)\*]'
def remove_special_characters(sentence):
    tmp = re.sub('’ ', ' ', sentence)
    tmp = re.sub("’$", '', tmp)
    tmp = re.sub('’', '\'', tmp)
    tmp = re.sub(chars_to_ignore_regex, '', tmp)
    sentence = irish_lower_sentence(tmp) + ' '
    return sentence
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the audio files as arrays
def speech_file_to_array_fn(batch):
    batch["sentence"] = remove_special_characters(batch["sentence"])
    speech_array, sampling_rate = torchaudio.load(batch["path"])
    batch["speech"] = resampler(speech_array).squeeze().numpy()
    return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
# Preprocessing the datasets.
# We need to read the audio 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: 43.7 %

Training

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

The script used for training can be found here