script/librispeech.py

import pandas as pd
from process_from_parquet import read_parquet_file, process_parquet_df, save_to_csv
from process_audio import process_audio_column


def process_partition(partition, process_row_with_params):
    """
    Process the partition after first row processing. 
    Covert the series result to dataframe to further processing for audio partition.
    
    """
    result = partition.apply(process_row_with_params, axis=1)
    field_name = ["path", "url" ,"type", "duration", "language", "transcript", "tag", "split", "license", "audio"]
    return pd.DataFrame(result.tolist(), columns=field_name)  # Convert Series to DataFrame

def _get_split(parquet_file):
    if "train" in parquet_file:
        return "train"
    elif "test" in parquet_file:
        return "test"
    elif "dev" in parquet_file:
        return "validation"
    else:
        return "train"

def process_row(row, parquet_file_name):
    """
    The function to process each row from dataframe.
    Return the metadata as dictionary. 
    
    """

    metadata = {}
  
    metadata["audio"] = row["audio"]
    metadata["url"] = f"https://huggingface.co/datasets/meetween/mumospee_librispeech/resolve/main/librispeech-parquet/{parquet_file_name}"
    metadata["transcript"] = row["text"]
    metadata["type"] = "audio"
    metadata["language"] = "en"
    metadata["tag"] = "Librispeech"
    metadata["split"] = _get_split(parquet_file_name)
    metadata["license"] = "CC-BY-4.0"

    return metadata

def main(config):
    parquet_df, file_name = read_parquet_file(config["parquet_file_path"], top=config["top"])

    processed_df = process_parquet_df(parquet_df=parquet_df,
                                     file_name=file_name,
                                     process_row_func=process_row,
                                     process_partition=process_partition)
    

    result_df = process_audio_column(processed_df)

    save_to_csv(result_df, final_path=config["final_path"])