3 people e61c85365b fix(backend): enable long-form Whisper transcription on PyTorch path ...

The PyTorch Whisper transcribe() called the HF processor without
truncation=False and model.generate() without return_timestamps=True.
With those defaults, WhisperFeatureExtractor silently truncates inputs
to 30 s (Whisper's native receptive field), so any dictation longer
than ~30 s lost its tail.

Setting truncation=False + padding="longest" + return_attention_mask=True
on the processor, then forwarding the attention mask plus
return_timestamps=True to generate(), flips HF Whisper into long-form
mode: autoregressive decoding over rolling 30 s windows.

Verified by round-tripping a 56.6 s Kokoro TTS sample through
/transcribe — full text returned including content past the 30 s mark;
previously the transcript was cut off roughly halfway through.

MLX backend (mlx_backend.py) is intentionally unchanged: mlx_audio.stt's
generate() already implements rolling-window long-form transcription
with condition_on_previous_text in the upstream library, so it does not
have the same bug. The HF-only kwargs added here would also break the
MLX call signature.

Co-Authored-By: Claude Opus 4.7 <[email protected]>

2026-10-04 00:01:34 +00:00

..

__init__.py

fix(kokoro): trim edges only, keep inter-segment gaps

2026-10-04 00:01:24 +00:00

base.py

refactor(generation): simplify post-generation cache cleanup

2026-10-04 00:01:18 +00:00

chatterbox_backend.py

fix(backend): prevent unbounded memory accumulation over consecutive TTS generations (#923)

2026-10-04 00:01:18 +00:00

chatterbox_turbo_backend.py

fix(backend): prevent unbounded memory accumulation over consecutive TTS generations (#923)

2026-10-04 00:01:18 +00:00

hume_backend.py

fix(tada): run voice-prompt encode under torch.inference_mode (#955)

2026-07-26 23:31:20 -07:00

kokoro_backend.py

fix(kokoro): trim edges only, keep inter-segment gaps

2026-10-04 00:01:24 +00:00

luxtts_backend.py

fix(backend): prevent unbounded memory accumulation over consecutive TTS generations (#923)

2026-10-04 00:01:18 +00:00

mlx_backend.py

fix(offline): patch transformers mistral-regex check to survive HF failures (#530)

2026-04-21 22:01:29 -07:00

pytorch_backend.py

fix(backend): enable long-form Whisper transcription on PyTorch path

2026-10-04 00:01:34 +00:00

qwen_custom_voice_backend.py

fix(backend): prevent unbounded memory accumulation over consecutive TTS generations (#923)

2026-10-04 00:01:18 +00:00

qwen_llm_backend.py

fix(offline): remove process-global offline guard from Qwen3 LLM loads (#924)

2026-07-20 18:47:10 -07:00