process_datasett.py

import csv
import json
import os
import wave

# Give you config paths carefully
INPUT_TSV = "input.tsv"   
AUDIO_DIR = "training_audio\\<audio>\\audio"
OUTPUT_DIR = "training_audio\\<audio\\transcripts"
SOURCE = "openslr"
GENDER = "female"
LANGUAGE = "assamese"

os.makedirs(OUTPUT_DIR, exist_ok=True)

def clean_text(text):
    text = text.replace("\\n", " ").strip()
    text = " ".join(text.split())
    return text

def get_wav_duration(path):
    try:
        with wave.open(path, 'r') as wf:
            frames = wf.getnframes()
            rate = wf.getframerate()
            return round(frames / float(rate), 3)
    except:
        return None

def sort_key(uid):
    """
    Example UID:
    train_assamesemale_00001.txt
    """
    try:
        num = uid.split("_")[-1].replace(".txt", "")
        return int(num)
    except:
        return 0


rows = []

with open(INPUT_TSV, "r", encoding="utf-8") as f:
    reader = csv.reader(f, delimiter="\t")

    for row in reader:
        if len(row) < 2:
            continue

        uid = row[0].strip()
        text = clean_text(row[1])

        rows.append([uid, text])

rows.sort(key=lambda x: sort_key(x[0]))

for uid, text in rows:

    base_id = uid.replace(".txt", "")

    audio_path = os.path.join(AUDIO_DIR, f"{base_id}.wav")
    duration = get_wav_duration(audio_path)

    data = {
        "id": base_id,
        "audio": audio_path,
        "transcript": text,
        "language": LANGUAGE,
        "gender": GENDER,
        "source": SOURCE,
        "duration": duration
    }

    json_path = os.path.join(OUTPUT_DIR, f"{base_id}.json")

    with open(json_path, "w", encoding="utf-8") as jf:
        json.dump(data, jf, ensure_ascii=False, indent=2)