openai/whisper · error · ValueError

Unsupported language: {language}

Error message

Unsupported language: {language}

What it means

get_tokenizer() accepts a language only if it is either a key of LANGUAGES (a code like 'en', 'de') or a key of TO_LANGUAGE_CODE (an English name like 'english', 'german', which it then converts). Anything else — before the multilingual branch even runs — raises ValueError('Unsupported language: ...').

Source

Thrown at whisper/tokenizer.py:380

        special_tokens=special_tokens,
    )


@lru_cache(maxsize=None)
def get_tokenizer(
    multilingual: bool,
    *,
    num_languages: int = 99,
    language: Optional[str] = None,
    task: Optional[str] = None,  # Literal["transcribe", "translate", None]
) -> Tokenizer:
    if language is not None:
        language = language.lower()
        if language not in LANGUAGES:
            if language in TO_LANGUAGE_CODE:
                language = TO_LANGUAGE_CODE[language]
            else:
                raise ValueError(f"Unsupported language: {language}")

    if multilingual:
        encoding_name = "multilingual"
        language = language or "en"
        task = task or "transcribe"
    else:
        encoding_name = "gpt2"
        language = None
        task = None

    encoding = get_encoding(name=encoding_name, num_languages=num_languages)

    return Tokenizer(
        encoding=encoding, num_languages=num_languages, language=language, task=task
    )

View on GitHub (pinned to 5f86d1d863)

Solutions

1. Normalize to the base two-letter code: language.split('-')[0].strip().lower() and verify against whisper.tokenizer.LANGUAGES
2. Pass the English display name ('german') only if spelled exactly as in TO_LANGUAGE_CODE
3. Validate user-supplied languages against whisper.tokenizer.LANGUAGES before calling transcribe()

Example fix

# before
result = whisper.transcribe(model, path, language="zh-CN")  # ValueError

# after
from whisper.tokenizer import LANGUAGES
lang = "zh-CN".split("-")[0].strip().lower()
assert lang in LANGUAGES, f"unsupported: {lang}"
result = whisper.transcribe(model, path, language=lang)

from whisper.tokenizer import LANGUAGES, TO_LANGUAGE_CODE

def normalize_language(raw: str) -> str:
    lang = raw.strip().lower()
    if lang in LANGUAGES:
        return lang
    if lang in TO_LANGUAGE_CODE:
        return TO_LANGUAGE_CODE[lang]
    base = lang.split("-")[0].split("_")[0]
    if base in LANGUAGES:
        return base
    raise ValueError(f"Unsupported language: {raw}")

def is_supported_language(v: str) -> bool:
    v = v.strip().lower()
    return v in LANGUAGES or v in TO_LANGUAGE_CODE

When it happens

Trigger: get_tokenizer(True, language='zh-CN') (locale form instead of 'zh'); language='eng' (ISO-639-2/b code); passing None-with-typo or a name with trailing whitespace ('english ') so neither dict matches; also raised via transcribe(model, path, language=<bad>).

Common situations: Feeding BCP-47 locale tags from web requests directly into language=; using 3-letter ISO codes from other datasets; uppercase already handled (.lower() is applied) but hyphens/whitespace are not.

Related errors

• Language {language} not found in tokenizer.
• {download_target} exists and is not a regular file
• Model {name} not found; available models = {available_models
• This model doesn't have language tokens so it can't perform
• This tokenizer does not have language token configured

AI-assisted analysis of openai/whisper@5f86d1d863 (2026-08-14). Data as JSON: /api/errors/0c09b5450a5dc463. Report an issue: GitHub.