#1504 numpy.core._exceptions._ArrayMemoryError: Unable to allocate 384. MiB for an array with shape (1, 125986, 400) and data

83.165* Posted at: 3 months ago 👁144

Traceback (most recent call last):
File "videotrans\process\_overall.py", line 57, in run
File "faster_whisper\transcribe.py", line 902, in transcribe
File "faster_whisper\feature_extractor.py", line 215, in call
File "faster_whisper\feature_extractor.py", line 189, in stft
File "numpy\fft\_pocketfft.py", line 409, in rfft
File "numpy\fft\_pocketfft.py", line 70, in _raw_fft
numpy.core._exceptions._ArrayMemoryError: Unable to allocate 384. MiB for an array with shape (1, 125986, 400) and data type float64

=
system:Windows-10-10.0.19044-SP0
version:v3.83
frozen:True
language:en
root_dir:C:/win-videotrans-v3.83

1 Replies

Floor #1 DeepSeek AI reply at 3 months ago

This error indicates insufficient GPU memory (VRAM) during the faster-whisper speech recognition process. The model is trying to allocate 384 MiB but failing. Here are the solutions:

Switch to a smaller model: In the speech recognition settings, change from large-v2/large-v3 to a smaller model like medium or small.
Enable CUDA optimization: In the main interface, ensure "CUDA加速" is checked if you have an NVIDIA GPU.
Adjust CUDA data type: Go to Menu → Tools → Advanced Options and set "CUDA数据类型" to int8 or float16 instead of float32 to reduce memory usage.
Reduce recognition accuracy settings: In the same Advanced Options, lower both "字幕识别准确度控制beam_size" and "字幕识别准确度控制best_of" to 1 or 2.
Close other GPU-intensive applications to free up VRAM.

If these steps don't resolve the issue, you may need to use CPU-only processing (uncheck "CUDA加速"), though this will be significantly slower.