import os import re import json import torch import argparse from functools import partial from dataclasses import dataclass, field from typing import Any, Dict, List, Optional, Union import numpy as np import pandas as pd from datasets import set_caching_enabled set_caching_enabled(False) from datasets import ( load_dataset, load_from_disk, load_metric,) from transformers import ( Wav2Vec2CTCTokenizer, Wav2Vec2FeatureExtractor, Wav2Vec2Processor, Wav2Vec2ForCTC, TrainingArguments, Trainer, ) import torchaudio def preprocess_data(example, tok_func = word_tokenize): example['sentence'] = ' '.join(tok_func(example['sentence'])) return example def speech_file_to_array_fn(batch, text_col="sentence", fname_col="path", resampling_to=16000): speech_array, sampling_rate = torchaudio.load(batch[fname_col]) resampler=torchaudio.transforms.Resample(sampling_rate, resampling_to) batch["speech"] = resampler(speech_array)[0].numpy() batch["sampling_rate"] = resampling_to batch["target_text"] = batch[text_col] return @dataclass class DataCollatorCTCWithPadding: """ Data collator that will dynamically pad the inputs received. Args: processor (:class:`~transformers.Wav2Vec2Processor`) The processor used for proccessing the data. padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`): Select a strategy to pad the returned sequences (according to the model's padding side and padding index) among: * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single sequence if provided). * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the maximum acceptable input length for the model if that argument is not provided. * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different lengths). max_length (:obj:`int`, `optional`): Maximum length of the ``input_values`` of the returned list and optionally padding length (see above). max_length_labels (:obj:`int`, `optional`): Maximum length of the ``labels`` returned list and optionally padding length (see above). pad_to_multiple_of (:obj:`int`, `optional`): If set will pad the sequence to a multiple of the provided value. This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >= 7.5 (Volta). """ processor: Wav2Vec2Processor padding: Union[bool, str] = True max_length: Optional[int] = None max_length_labels: Optional[int] = None pad_to_multiple_of: Optional[int] = None pad_to_multiple_of_labels: Optional[int] = None def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]: # split inputs and labels since they have to be of different lenghts and need # different padding methods input_features = [{"input_values": feature["input_values"]} for feature in features] label_features = [{"input_ids": feature["labels"]} for feature in features] batch = self.processor.pad( input_features, padding=self.padding, max_length=self.max_length, pad_to_multiple_of=self.pad_to_multiple_of, return_tensors="pt", ) with self.processor.as_target_processor(): labels_batch = self.processor.pad( label_features, padding=self.padding, max_length=self.max_length_labels, pad_to_multiple_of=self.pad_to_multiple_of_labels, return_tensors="pt", ) # replace padding with -100 to ignore loss correctly labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100) batch["labels"] = labels return batch def main(): parser = argparse.ArgumentParser() parser.add_argument("--pre_trained_model", default='', type=str, help='Local path to pre-trained wav2vec2 model') parser.add_argument("--train_file_path", default='', type=str, help='Local path to train file') parser.add_argument("--valid_file_path", default='', type=str, help='Local path to valid file') parser.add_argument("--warmup_steps", default=20000, type=int, help='') parser.add_argument("--learning_rate", default=3e-5, type=float, help='') args = parser.parse_args() def prepare_dataset(batch): # check that all files have the correct sampling rate # assert ( # len(set(batch["sampling_rate"])) == 1 # ), f"Make sure all inputs have the same sampling rate of {processor.feature_extractor.sampling_rate}." batch["input_values"] = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0]).input_values with processor.as_target_processor(): batch["labels"] = processor(batch["target_text"]).input_ids return def compute_metrics(pred, processor, metric): pred_logits = pred.predictions pred_ids = np.argmax(pred_logits, axis=-1) pred.label_ids[pred.label_ids == -100] = processor.tokenizer.pad_token_id pred_str = processor.batch_decode(pred_ids) # we do not want to group tokens when computing the metrics label_str = processor.batch_decode(pred.label_ids, group_tokens=False) wer = cer_metric.compute(predictions=pred_str, references=label_str) return {"cer": cer} # load dataset print('Loading dataset....') datasets = load_dataset('csv', name='cn', data_files={'train': args.train_file_path, 'valid': args.valid_file_path}, cache_dir='/path/to/csv') datasets = datasets.map(preprocess_data) dataset_train = datasets['train'] dataset_valid = datasets['valid'] dataset_train = dataset_train.map(speech_file_to_array_fn, remove_columns=dataset_train.column_names, cache_file_name='/path/to/cache/of/train/speech/file') dataset_valid = dataset_valid.map(speech_file_to_array_fn, remove_columns=dataset_valid.column_names, cache_file_name='/path/to/cache/of/valid/speech/file') print('Tokenization') tokenizer = Wav2Vec2CTCTokenizer.from_pretrained(args.pre_trained_model) print('Feature extracting....') feature_extractor = Wav2Vec2FeatureExtractor(args.pre_trained_model) processor = Wav2Vec2Processor(feature_extractor=feature_extractor, tokenizer=tokenizer) dataset_train = dataset_train.map(prepare_dataset, remove_columns=dataset_train.column_names, batched=True, load_from_cache_file=True, cache_file_name='/path/to/train') dataset_valid = dataset_valid.map(prepare_dataset, remove_columns=dataset_valid.column_names, batched=True, load_from_cache_file=True, cache_file_name='/path/to/valid') data_collator = DataCollatorCTCWithPadding(processor=processor, padding=True) wer_metric = load_metric("cer") # create model model = Wav2Vec2ForCTC.from_pretrained( args.pre_trained_model, vocab_size=len(processor.tokenizer) ) model.freeze_feature_extractor() training_args = TrainingArguments( output_dir="/path/to/output", group_by_length=True, per_device_train_batch_size=3, gradient_accumulation_steps=1, per_device_eval_batch_size=1, metric_for_best_model='cer', evaluation_strategy="steps", eval_steps=15000, logging_strategy="steps", logging_steps=15000, save_strategy="steps", save_steps=15000, num_train_epochs=100, fp16=True, learning_rate=args.learning_rate, warmup_steps=args.warmup_steps, save_total_limit=3, report_to="tensorboard" ) print('Training model....') # Train trainer = Trainer( model=model, data_collator=data_collator, args=training_args, compute_metrics=partial(compute_metrics, metric=cer_metric, processor=processor), train_dataset=dataset_train, eval_dataset=dataset_valid, tokenizer=processor.feature_extractor, ) trainer.train()