whisper_optimized_processing.py

import os
from datasets import Dataset, DatasetDict, load_dataset
from datasets.features import Audio
import pandas as pd
import numpy as np
from tqdm import tqdm

# Function to load your custom dataset
def load_custom_dataset(data_dir):
    data = {
        "audio": [],
        "text": []
    }

    wav_dir = os.path.join(data_dir, 'wav')
    txt_dir = os.path.join(data_dir, 'transcription')

        # Assuming filenames in 'wav' and 'txt' match
    for wav_file in os.listdir(wav_dir):
            if wav_file.endswith('.wav'):
                txt_file = wav_file.replace('.wav', '.txt')
                wav_path = os.path.join(wav_dir, wav_file)
                txt_path = os.path.join(txt_dir, txt_file)

                # Read the transcription text
                with open(txt_path, 'r', encoding='utf-8') as f:
                    transcription = f.read().strip()

                # Append to the dataset
                data["audio"].append(wav_path)
                data["text"].append(transcription)

    # Create a pandas dataframe
    df = pd.DataFrame(data)

    # Convert to a Hugging Face dataset
    dataset = Dataset.from_pandas(df)

    # Define the audio feature (for .wav files)
    dataset = dataset.cast_column("audio", Audio(sampling_rate=16_000))  # Adjust the sampling rate if needed

    return dataset

# Load your custom dataset
custom_train_dataset = load_custom_dataset("./")

# Load Common Voice test set (Malayalam)
common_voice_test = load_dataset("mozilla-foundation/common_voice_11_0", "ml", split="test", trust_remote_code=True)

common_voice_test = common_voice_test.select_columns(["audio", "sentence"])

# Combine them into a DatasetDict
dataset_dict = DatasetDict({
    "train": custom_train_dataset,
    "test": common_voice_test
})

# Now you have the `dataset_dict` with your custom train set and the Common Voice test set
print(dataset_dict)

from transformers import WhisperFeatureExtractor

feature_extractor = WhisperFeatureExtractor.from_pretrained("openai/whisper-small")

from transformers import WhisperTokenizer

tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-small", language="Malayalam", task="transcribe")

from transformers import WhisperProcessor

processor = WhisperProcessor.from_pretrained("openai/whisper-small", language="Malayalam", task="transcribe")

print(dataset_dict['train'][0])

import gc  # for garbage collection

def prepare_dataset(batch):
    # Prepare input features for each audio file in the batch
    audio_arrays = [item["array"] for item in batch["audio"]]
    sampling_rates = [item["sampling_rate"] for item in batch["audio"]]

    # Extract features for each audio sample
    features = []
    for audio, sr in zip(audio_arrays, sampling_rates):
        feature = feature_extractor(audio, sampling_rate=sr).input_features[0]
        feature = np.array(feature, dtype=np.float16)
        features.append(feature)

        # Free memory after each feature extraction
        del audio  # Remove reference to the audio array
        del sr
        gc.collect()  # Trigger garbage collection to free memory

    # Store features in batch
    batch["input_features"] = features

    # Encode target text to label ids
    # Consider using a tokenizer with padding strategy (e.g., `padding="max_length"` or `padding="longest"`)
    batch["labels"] = tokenizer(batch["text"], padding="longest", truncation=True).input_ids

    return batch

# Function to process and save dataset in batches
def process_and_save_in_batches(dataset, batch_size=1000, save_path="processed_dataset"):
    # Create an empty list to store the processed batches
    all_processed = []

    # Loop through the dataset in chunks
    for start_idx in range(0, len(dataset), batch_size):
        # Get the batch slice
        batch = dataset[start_idx:start_idx+batch_size]
        batch = Dataset.from_dict(batch)
        # Apply the processing function to the batch
        processed_batch = batch.map(
            prepare_dataset,
            remove_columns=dataset.column_names,
            batched=True,
            batch_size=batch_size,
            num_proc = None,
        )
        print(f"Batch {start_idx} done")
        # Append the processed batch to the list
        all_processed.append(processed_batch)
        
        # Clear memory after processing each batch
        del batch  # Remove reference to the batch
        gc.collect()  # Trigger garbage collection

        # Save each processed batch to disk
        processed_batch.save_to_disk(os.path.join(save_path, f"batch_{start_idx // batch_size}"))
        del processed_batch  # Free memory after saving the batch
        gc.collect()

    # Optionally, if you want to save the whole dataset in one file at the end
    # You can merge all processed batches (not recommended for large datasets)
    final_dataset = concatenate_datasets(all_processed)
    final_dataset.save_to_disk(save_path)

# Process and save the dataset in batches
process_and_save_in_batches(dataset_dict['train'], batch_size=1000, save_path="processed_dataset")