fix_languages.py

# %%
from pathlib import Path
import json
import logging

import datasets
import lingua

# %%
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)

# %%
language_detector = lingua.LanguageDetectorBuilder.from_languages(
    lingua.Language.ENGLISH,
    lingua.Language.SPANISH,
    lingua.Language.GERMAN,
    lingua.Language.FRENCH,
).build()

# %%
fixes = []
language_cache = {}
for path in sorted(Path(".").glob("**/metadata.jsonl")):
    dataset = datasets.load_dataset(
        "json",
        data_files=str(path),
        split="train",
    )
    for record in dataset:
        cache_key = (record["song_name"], record["text"].strip())
        if cache_key in language_cache:
            language = language_cache[cache_key]
        else:
            language = record["language"]
            detected_languages = {
                x.language.iso_code_639_1.name.lower(): x.value
                for x in language_detector.compute_language_confidence_values(
                    record["text"]
                )
            }
            detected_languages = {
                lg: p for lg, p in detected_languages.items() if p > 0.4
            }
            if record["language"] not in detected_languages:
                print(
                    f"{record['song_name']}: Detected languages {detected_languages} instead of "
                    f"{repr(language)} for line:\n{record['text'].strip()}"
                )
                language = input(f"Enter language [{language}]: ").strip() or language
                print()

            language_cache[cache_key] = language

        if language != record["song_language"]:
            fixes.append(
                {
                    "file_name": record["file_name"],
                    "text": record["text"],
                    "language": language,
                }
            )

Path("language_fixes.jsonl").write_text(
    "\n".join(json.dumps(f) for f in fixes) + "\n", encoding="utf-8"
)