Skip to content

The open-source AI voice studio. Clone, dictate, create.

fix(backend): pass language/task directly to Whisper generate

nox7575a65e7f84
diff --git a/backend/backends/pytorch_backend.py b/backend/backends/pytorch_backend.py
index 3e1db396..012fd85e 100644
--- a/backend/backends/pytorch_backend.py
+++ b/backend/backends/pytorch_backend.py
@@ -358,15 +358,17 @@ class PyTorchSTTBackend:
)
inputs = inputs.to(self.device)
- # Generate transcription
- # If language is provided, force it; otherwise let Whisper auto-detect
+ # Generate transcription.
+ # If language is provided, force it; otherwise let Whisper
+ # auto-detect. Pass language/task directly to generate() instead
+ # of building forced_decoder_ids — get_decoder_prompt_ids defaults
+ # to no_timestamps=True, which injects <|notimestamps|> and
+ # disables the timestamp tokens that return_timestamps=True (and
+ # therefore long-form decoding) depend on.
generate_kwargs = {}
if language:
- forced_decoder_ids = self.processor.get_decoder_prompt_ids(
- language=language,
- task="transcribe",
- )
- generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
+ generate_kwargs["language"] = language
+ generate_kwargs["task"] = "transcribe"
with torch.no_grad():
predicted_ids = self.model.generate(