openai/whisper · error · ValueError

This model doesn't have language tokens so it can't perform

Error message

This model doesn't have language tokens so it can't perform lang id

What it means

whisper.detect_language() needs a tokenizer whose prompt sequence contains a language token (<|xx|> after <|startoftranscript|>). English-only models ('.en' checkpoints, or get_tokenizer(multilingual=False)) have no language tokens, so the check tokenizer.language_token not in sot_sequence raises ValueError before any inference.

Source

Thrown at whisper/decoding.py:42

    of the most probable language tokens and the probability distribution over all language tokens.
    This is performed outside the main decode loop in order to not interfere with kv-caching.

    Returns
    -------
    language_tokens : Tensor, shape = (n_audio,)
        ids of the most probable language tokens, which appears after the startoftranscript token.
    language_probs : List[Dict[str, float]], length = n_audio
        list of dictionaries containing the probability distribution over all languages.
    """
    if tokenizer is None:
        tokenizer = get_tokenizer(
            model.is_multilingual, num_languages=model.num_languages
        )
    if (
        tokenizer.language is None
        or tokenizer.language_token not in tokenizer.sot_sequence
    ):
        raise ValueError(
            "This model doesn't have language tokens so it can't perform lang id"
        )

    single = mel.ndim == 2
    if single:
        mel = mel.unsqueeze(0)

    # skip encoder forward pass if already-encoded audio features were given
    if mel.shape[-2:] != (model.dims.n_audio_ctx, model.dims.n_audio_state):
        mel = model.encoder(mel)

    # forward pass using a single token, startoftranscript
    n_audio = mel.shape[0]
    x = torch.tensor([[tokenizer.sot]] * n_audio).to(mel.device)  # [n_audio, 1]
    logits = model.logits(x, mel)[:, 0]

    # collect detected languages; suppress all non-language tokens
    mask = torch.ones(logits.shape[-1], dtype=torch.bool)

View on GitHub (pinned to 5f86d1d863)

Solutions

1. Use a multilingual model: whisper.load_model('base') instead of 'base.en'
2. Skip detect_language() for English-only models and hardcode language='en' in transcribe()
3. Gate the call on model.is_multilingual before invoking detect_language()

Example fix

# before
model = whisper.load_model("base.en")
lang, _ = whisper.detect_language(model, mel)  # ValueError

# after
model = whisper.load_model("base")
lang, _ = whisper.detect_language(model, mel)
# or keep .en model and skip detection:
# result = whisper.transcribe(model, path, language="en")

def supports_lang_id(model) -> bool:
    return model.is_multilingual

from whisper import Whisper

def can_detect_language(model: Whisper) -> bool:
    return bool(getattr(model, "is_multilingual", False))

try:
    _, probs = whisper.detect_language(model, mel)
except ValueError as e:
    if "language tokens" in str(e):
        probs = {"en": 1.0}  # English-only model: assume en
    else:
        raise

When it happens

Trigger: Calling whisper.detect_language(model) on a model loaded as 'base.en'/'large-v2.en'; or building DecodingOptions/Tokenizer manually with multilingual=False and passing it to detect_language; the code path builds a default tokenizer from model.is_multilingual and it comes back English-only.

Common situations: Switching a pipeline from 'base' to 'base.en' for speed and forgetting a detect_language() call; fine-tuned checkpoints whose dims.n_vocab reflect the English vocabulary; code that assumes every model supports lang id.

Related errors

• This tokenizer does not have language token configured
• Language {language} not found in tokenizer.
• Unsupported language: {language}

AI-assisted analysis of openai/whisper@5f86d1d863 (2026-08-14). Data as JSON: /api/errors/70d969e57569f265. Report an issue: GitHub.