search

jianfch / stable-ts

Transcription, forced alignment, and audio indexing with OpenAI's Whisper
MIT License
1.59k stars 176 forks source link

RuntimeError: quantile() input tensor is too large #388

Closed xu2k closed 2 months ago

xu2k commented 3 months ago

Traceback (most recent call last): File "C:\Users\sgwli\OneDrive\바탕 화면\stable-ts-2.0.0\run.py", line 3, in result = model.transcribe('audio.mp3') ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\sgwli\OneDrive\바탕 화면\stable-ts-2.0.0\stable_whisper\whisper_word_level.py", line 359, in transcribe_stable ts_token_mask = wav2mask(audio, q_levels=q_levels, k_size=k_size) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\sgwli\OneDrive\바탕 화면\stable-ts-2.0.0\stable_whisper\stabilization.py", line 200, in wav2mask loudness_tensor = audio2loudness(standardize_audio(audio)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\sgwli\OneDrive\바탕 화면\stable-ts-2.0.0\stable_whisper\stabilization.py", line 129, in audio2loudness audio_tensor = audio_tensor / min(1., (audio_tensor.quantile(0.999, dim=-1) * 1.75)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: quantile() input tensor is too large

jianfch commented 3 months ago

You are running an old version. Update Stable-ts and it should be fixed.