stats.py

from typing import List, Literal

import numpy as np
import torchaudio
from datasets import concatenate_datasets, load_dataset
from pandas import concat, read_csv
from tqdm import tqdm


def get_audio_length(file_path: str) -> float:
    """
    計錄音總長度
    """
    metadata = torchaudio.info(file_path)
    return metadata.num_frames / metadata.sample_rate


def get_info(subsets: List[str] | Literal["saamgwokjinji", "seoiwuzyun"]) -> None:
    """
    計音頻長度、字數、平均字數、中位數字數、覆蓋字數、平均語速
    """
    if not isinstance(subsets, list):
        subsets = [subsets]

    datasets = []
    for subset in subsets:
        dataset = load_dataset('audiofolder', data_dir=f'./opus/{subset}', split='train')
        datasets.append(dataset)

    if len(datasets) > 1:
        dataset = concatenate_datasets(datasets)
    else:
        dataset = datasets[0]

    # List to store individual audio durations
    durations = []
    total_duration = 0

    for item in tqdm(dataset):
        file_path = item['audio']['path']
        duration = get_audio_length(file_path)
        durations.append(duration)
        total_duration += duration

    # Calculate statistics
    min_duration = min(durations)
    max_duration = max(durations)
    avg_duration = np.mean(durations)
    median_duration = np.median(durations)

    # Print results
    print(f"Statistics for: {' & '.join(subsets)}")
    print(f"Total audio duration: {total_duration / 3600:.2f} hours")
    print(f"Total audio duration: {total_duration / 60:.2f} minutes")
    print(f"Minimum audio duration: {min_duration:.3f} seconds")
    print(f"Maximum audio duration: {max_duration:.3f} seconds")
    print(f"Average audio duration: {avg_duration:.3f} seconds")
    print(f"Median audio duration: {median_duration:.3f} seconds")

    # Concatenate metadata DataFrames
    metadata_list = []
    for subset in subsets:
        metadata = read_csv(f'./opus/{subset}/metadata.csv')
        metadata_list.append(metadata)

    if len(metadata_list) > 1:
        metadata = concat(metadata_list)
    else:
        metadata = metadata_list[0]

    # Calculate total number of characters
    total_characters = metadata['transcription'].str.len().sum()

    # Calculate mean number of characters
    mean_characters = metadata['transcription'].str.len().mean()

    # Calculate median number of characters
    median_characters = metadata['transcription'].str.len().median()

    # Get unique characters
    unique_characters = set(''.join(metadata['transcription']))

    # Print results
    print(f"Total characters: {total_characters}")
    print(f"Mean characters per transcription: {mean_characters:.2f}")
    print(f"Median characters per transcription: {median_characters}")
    print(f"Number of unique characters: {len(unique_characters)}")
    print(f"Unique characters: {''.join(sorted(unique_characters))}")
    print(f"Average speech rate: {total_characters / total_duration:.2f} characters per second")
    print("-" * 20)


if __name__ == '__main__':
    get_info('saamgwokjinji')
    get_info('seoiwuzyun')
    get_info(['saamgwokjinji', 'seoiwuzyun'])