dataset_download.py

from datasets import load_dataset, Audio
import soundfile as sf, os, re, neologdn, librosa
from tqdm import tqdm
import shutil

def have(a):
    return a is not None

def aorb(a, b):
    return a if have(a) else b

dataset = load_dataset("Sin2pi/JA_audio_JA_text_180k_samples", trust_remote_code=True)["train"].filter(lambda sample: bool(sample["sentence" if "sentence" in sample else aorb("text", "transcription")]))
name = "JA_audio_JA_text_180k"

ouput_dir = "./datasets/"
out_file = 'metadata.csv'
os.makedirs(ouput_dir + name, exist_ok=True)
folder_path = ouput_dir + name

top_db=30

def is_silent(mp3_file, threshold=0.025):
    if not os.path.exists(mp3_file):
        return True 
    y, sr = librosa.load(mp3_file, sr=None)
    rms = librosa.feature.rms(y=y)[0]
    return all(value < threshold for value in rms)

def remove_silence(input_file, output_file, top_db=top_db):
        y, sr = sf.read(input_file)
        intervals = librosa.effects.split(y, top_db=top_db)
        y_trimmed = []
        for start, end in intervals:
            y_trimmed.extend(y[start:end])          
        if not os.path.exists(output_file):    
            sf.write(output_file, y_trimmed, sr)
            with open(csv_file2, "a", encoding='utf-8') as f:
                file_name = os.path.basename(output_file)
                f.write(file_name + "\n")

def process_directory(input_dir, output_dir, top_db=top_db):
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)
    if not os.path.exists(removed_dir):
        os.makedirs(removed_dir)
    open(csv_file, 'w', encoding='utf-8').close()
    open(csv_file2, 'w', encoding='utf-8').close()
    
    for filename in os.listdir(input_dir):
        if filename.endswith(".mp3"):
            input_file = os.path.join(input_dir, filename)
            output_file = os.path.join(output_dir, filename)
            removed_file = os.path.join(removed_dir, filename)
            
            if not os.path.exists(output_file):
                remove_silence(input_file, output_file, top_db)

            if os.path.exists(output_file) and is_silent(output_file):
                with open(csv_file, "a", encoding='utf-8') as f: 
                    f.write(os.path.basename(output_file) + "\n") 
                shutil.move(output_file, removed_file) 
            
            if os.path.exists(input_file):
                os.remove(input_file)

input_dir = folder_path
output_dir = folder_path + "/trimmed/"
removed_dir = folder_path + "/removed/"
csv_file = folder_path + "/removed.csv"   
csv_file2 = folder_path + "/not_removed.csv"  

min_char = 4
max = 20.0
min = 1.0

char = '[　０１２３４５６７８９ａｂｃｄｅｆｇｈｉｊｋｌｍｎｏｐｑｒｓｔｕｖｗｘｙｚＡＢＣＤＥＦＧＨＩＪＫＬＭＮＯＰＱＲＳＴＵＶＷＸＹＺabcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890♬♪♩♫]'
special_characters = '[“%‘”~゛＃＄％＆（）＊＋：；〈＝〉＠＾＿｛｜｝～"█』『.;:<>_()*&^$#@`, ]'

# dataset = dataset.cast_column("file_url", datasets.Audio())
dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
sentence_map = {}

open(os.path.join(folder_path, out_file), 'w', encoding='utf-8').close()

for i, sample in tqdm(enumerate(dataset)):
    if sample["sentence"] != "":
        audio_sample_name = name + f'_{i}.mp3' 
        audio_path_original = os.path.join(folder_path, audio_sample_name)

        patterns = [(r"…",'。'), (r"!!",'!'), (special_characters,""), (r"\s+", "")]
        for pattern, replace in patterns:

            sample["sentence"] = re.sub(pattern, replace, sample["sentence"])
        sample["sentence"] = (neologdn.normalize(sample["sentence"], repeat=1))
        if sample["sentence"][-1] not in ["!", "?", "。"]:
            sample["sentence"] += "。"
        
        sentence_length = len(sample["sentence"])
        audio_length = len(sample['file_url' if "file_url" in sample else "audio"]["array"]) / sample['file_url' if "file_url" in sample else "audio"]["sampling_rate"]

        if max > audio_length > min and not re.search(char, sample["sentence"]) and sentence_length > min_char and bool(sample["sentence"]):
            if not os.path.exists(audio_path_original):
                sf.write(audio_path_original, sample['file_url' if "file_url" in sample else "audio"]["array"], sample['file_url' if "file_url" in sample else "audio"]["sampling_rate"])
            sentence_map[audio_sample_name] = sample['sentence']
            
print(f"Downloaded {len(sentence_map)} audio files to {folder_path}. Starting silence trimming...")
process_directory(input_dir, output_dir)
print(f"Silence trimming complete. Trimmed files are in {output_dir}, silent files moved to {removed_dir}.")

print(f"Generating final metadata.csv in {folder_path}...")
with open(csv_file2, 'r', encoding='utf-8') as f_not_removed:
    for line in f_not_removed:
        trimmed_filename = line.strip() 
        
        if trimmed_filename in sentence_map:
            sentence = sentence_map[trimmed_filename]
            with open(os.path.join(folder_path, out_file), 'a', encoding='utf-8') as transcription_file:
                transcription_file.write(trimmed_filename + ",") 
                transcription_file.write(sentence)
                transcription_file.write('\n')
print(f"Metadata.csv generated for {os.path.join(folder_path, out_file)}.")