data_preprocessing.py 文件源码

def create_csv(data_dir):
    '''
    Generates CSV file (as required by DeepSpeech_RHL.py) in the given dir.

    Args:
        data_dir:   Directory where all .wav files and 
                    their associated timescripts are stored.

    '''

    # Get all audio and transcript file paths.
    audio_file_paths = sorted(glob.glob(data_dir + "*.wav"))
    transcript_file_paths = sorted(glob.glob(data_dir + "*.txt"))

    audio_file_sizes = []
    transcripts = []

    for x, y in zip(audio_file_paths, transcript_file_paths):
        with open(y, "rb") as f:
            transcripts.append(f.read())

        # Get file size.
        metadata = os.stat(x)
        audio_file_sizes.append(metadata.st_size)

    # Create pandas dataframe
    df = pandas.DataFrame(columns=["wav_filename", "wav_filesize", "transcript"])
    df["wav_filename"] = audio_file_paths
    df["wav_filesize"] = audio_file_sizes
    df["transcript"] = transcripts

    df.to_csv(data_dir + "data.csv", sep=",", index=None) # Save CSV