clean_lyrics.py

"""
clean_lyrics.py - Cleans raw scraped lyrics JSONL into training-ready JSONL.

Usage:
    python3 clean_lyrics.py
    python3 clean_lyrics.py --input my_raw.jsonl
    python3 clean_lyrics.py --no-training
"""

import json
import re
import os
import argparse

RAW_FILE      = "lyrics_raw.jsonl"
CLEAN_FILE    = "lyrics_clean.jsonl"
TRAINING_FILE = "sinhala_songs_training.jsonl"

BOILERPLATE = [
    'ගී පද',
    'ගීතයේ පද පෙළ',
    'මෙම වෙබ් අඩවියේ',
    'ඔව්, මෙම',
]

def is_sinhala(text):
    return any(0x0D80 <= ord(c) <= 0x0DFF for c in text)

def sinhala_char_count(text):
    return sum(1 for c in text if 0x0D80 <= ord(c) <= 0x0DFF)

def clean_lyrics(text):
    # 1. Remove boilerplate lines
    lines = text.split('\n')
    lines = [l for l in lines if not any(m in l for m in BOILERPLATE)]
    text = '\n'.join(lines)

    # 2. Remove URL lines and non-Sinhala lines
    lines = text.split('\n')
    lines = [l for l in lines if 'http' not in l and 'www.' not in l]
    lines = [l for l in lines if is_sinhala(l) or l.strip() == '']

    # 3. Remove trailing bare title line (short line with no verse markers)
    while lines:
        last = lines[-1].strip().rstrip('\r')
        if last == '':
            lines.pop()
            continue
        if len(last) < 50 and '...' not in last and '//' not in last:
            lines.pop()
        else:
            break

    # 4. Collapse 3+ blank lines into one
    text = '\n'.join(lines)
    text = re.sub(r'\n{3,}', '\n\n', text)
    return text.strip()

def to_training_examples(item):
    content = item['content']
    title   = item['title']
    artist  = item.get('artist', '')

    instructions = [
        'සිංහල ආදර ගීතයක් ලියන්න',
        'Give me a Sinhala love song',
        'Create a meaningful Sinhala love song',
        'සිංහල ගීතයක් ලියන්න',
        "'" + title + "' වැනි සිංහල ගීතයක් ලියන්න",
    ]
    if artist and artist != 'Unknown':
        instructions.append(artist + ' ගේ ශෛලියෙන් සිංහල ගීතයක් ලියන්න')

    examples = []
    for inst in instructions:
        prompt = (
            'Below is an instruction that describes a task. '
            'Write a response that appropriately completes the request.\n\n'
            '### Instruction:\n' + inst + '\n\n'
            '### Response:\n' + content
        )
        examples.append({
            'text': prompt, 'instruction': inst, 'output': content,
            'source': item.get('source', ''), 'type': item.get('type', 'song'),
            'title': title,
        })
    return examples

def main(raw_file, clean_file, training_file, skip_training):
    if not os.path.exists(raw_file):
        print('Input file not found: ' + raw_file)
        return

    total = kept = skipped = training_count = 0
    f_train = None if skip_training else open(training_file, 'w', encoding='utf-8')

    with open(raw_file, 'r', encoding='utf-8') as f_in, \
         open(clean_file, 'w', encoding='utf-8') as f_clean:

        for line in f_in:
            line = line.strip()
            if not line:
                continue
            total += 1
            try:
                item = json.loads(line)
            except json.JSONDecodeError:
                print('  Skipping malformed JSON on line ' + str(total))
                skipped += 1
                continue

            cleaned = clean_lyrics(item.get('content', ''))
            if sinhala_char_count(cleaned) < 50:
                skipped += 1
                print('  Skipped (too short): ' + item.get('title', '')[:50])
                continue

            item['content'] = cleaned
            f_clean.write(json.dumps(item, ensure_ascii=False) + '\n')
            kept += 1

            if f_train:
                for ex in to_training_examples(item):
                    f_train.write(json.dumps(ex, ensure_ascii=False) + '\n')
                    training_count += 1

    if f_train:
        f_train.close()

    print('\n' + '=' * 50)
    print('  CLEANING COMPLETE')
    print('=' * 50)
    print('  Input songs:       ' + str(total))
    print('  Kept:              ' + str(kept))
    print('  Skipped:           ' + str(skipped))
    print('  Clean JSONL:       ' + clean_file)
    if not skip_training:
        print('  Training examples: ' + str(training_count) + '  ->  ' + training_file)
    print('=' * 50)

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--input',       default=RAW_FILE)
    parser.add_argument('--output',      default=CLEAN_FILE)
    parser.add_argument('--training',    default=TRAINING_FILE)
    parser.add_argument('--no-training', action='store_true')
    args = parser.parse_args()
    main(args.input, args.output, args.training, args.no_training)