Baike.dev
All toolsAI codingTrendingOpen sourceNewsSubmit
Log in
Back to tool/Back to issues
#1342·faster-whisper

Divide by zero crash on Windows for specific audio files when word level timestamps are enabled with faster-whisper-large-v2 model

Author: idle-matrix-7890Created Aug 15, 2025Updated Jun 12, 2026

I'm seeing an issue when using faster-whisper to transcribe certain audio files where an error takes down the whole process and cannot be caught in Python. I've created an example repo that shows the issue though I'm unable to share the specific audio that causes the crash as it is customer data, we are still trying different options but it seems that the 'word_timestamps' setting and the model being used have an impact on whether the error appears - unfortunately word level timestamps are required for our use case and we don't have a free choice of models as a lot of the audio we process is low quality and accuracy is critical so I'm hoping to find alternative options I can try to resolve this.

Almost all audio files work absolutely fine with our current setup, but we have seen a couple like this where processing them kills the process. We haven't been able to find anything wrong with the files, they playback perfectly fine without errors and don't cause issues in any of our other software. The audio files we use are all '.wav' files using PCM S16 LE so there isn't any variation in formats or codecs.

Some general information about the environment:

OS: Windows Server 2019 Standard v1809 (also tested on Windows 11 using CPU transcription) GPU: NVIDIA L4 24GB Model: https://huggingface.co/Systran/faster-whisper-large-v2 NVIDIA CUDA Version: CUDA 12.3 installed, with the following libraries also accessible via the PATH environment variable - cublas64_12.dll (6.14.11.1234) - cublasLt64_12.dll (6.14.11.1234) - cudnn_cnn64_9.dll (9.10.2.21) - cudnn_engines_precompiled64_9.dll (9.10.2.21) - cudnn_engines_runtime_compiled64_9.dll (9.10.2.21) - cudnn_graph64_9.dll - cudnn_heuristic64_9.dll - cudnn_ops64_9.dll (9.10.2.21) - cudnn64_9.dll (9.10.2.21) Python: 3.10

nvidia-smi.exe output:

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 546.12                 Driver Version: 546.12       CUDA Version: 12.3     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                     TCC/WDDM  | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA L4                    TCC   | 00000000:03:00.0 Off |                    0 |
| N/A   73C    P0              33W /  72W |   4461MiB / 23034MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

An example of the output we see using the example repo when the error occurs is (paths sanitized for confidentiality):

C:\faster-whisper-divide-by-zero-issue\venv\Scripts\python.exe C:\faster-whisper-divide-by-zero-issue\main.py "C:\<PATH_TO_MODEL_FOLDER>" C:\<AUDIO_FILE>.wav 
INFO:faster_whisper:Processing audio with duration 04:04.692
INFO:faster_whisper:Detected language 'en' with probability 1.00
DEBUG:faster_whisper:Processing segment at 00:00.000
DEBUG:faster_whisper:Processing segment at 00:27.760
DEBUG:faster_whisper:Processing segment at 00:54.800
DEBUG:faster_whisper:Processing segment at 01:22.860
DEBUG:faster_whisper:Processing segment at 01:47.180
DEBUG:faster_whisper:Processing segment at 02:15.060
DEBUG:faster_whisper:Processing segment at 02:40.860
DEBUG:faster_whisper:Processing segment at 03:06.120
DEBUG:faster_whisper:Processing segment at 03:34.680
DEBUG:faster_whisper:Processing segment at 04:04.680
DEBUG:faster_whisper:Compression ratio threshold is not met with temperature 0.0 (13.250000 > 2.400000)
DEBUG:faster_whisper:Compression ratio threshold is not met with temperature 0.2 (13.250000 > 2.400000)
DEBUG:faster_whisper:Compression ratio threshold is not met with temperature 0.4 (13.250000 > 2.400000)
DEBUG:faster_whisper:Compression ratio threshold is not met with temperature 0.6 (13.250000 > 2.400000)
DEBUG:faster_whisper:Compression ratio threshold is not met with temperature 0.8 (5.176471 > 2.400000)

Process finished with exit code -1073741676 (0xC0000094)

Source: SYSTRAN/faster-whisper

View original on GitHubView discussion on GitHub