whisper takes a long time to do nothing on my media server

[garydale]

I'm running Debian/Trixie on an AMD64 system. I'm using an AMD Ryzen 3 1200 Quad-Core Processor with 16G of RAM and 16G of swap space. Free shows I'm using less than half of that when running whisper. My video card is ancient since I usually run the server headless. However it seemed like a good place to run whisper to generate subtitles.

Unfortunately, when I try, using the same scripts that worked on my workstation (same OS but faster processor and slightly better video card (I'm not a gamer so I never bothered getting a gamer video card), I find nothing is generated, Capturing the error output, I get this:

File "/home/garydale/.venv/lib/python3.13/site-packages/whisper/transcribe.py", line 615, in cli
result = transcribe(model, audio_path, temperature=temperature, **args)
File "/home/garydale/.venv/lib/python3.13/site-packages/whisper/transcribe.py", line 482, in transcribe
print(make_safe(line))
~~~~~^^^^^^^^^^^^^^^^^
OSError: [Errno 5] Input/output error
Exception ignored on flushing sys.stdout:
OSError: [Errno 5] Input/output error

The transcribe.py error matches the symptoms but I don't understand why I'm getting here and not on my other machine (which mainly uses a better processor but is otherwise similar).

Any ideas on how to fix this?

[shiv669]

Hey, this isn’t a Whisper bug. It’s your environment indicating there’s a problem writing to stdout. The OSError: [Errno 5] Input/output error usually occurs in headless systems when there’s no proper TTY attached or when stdout closes unexpectedly. Since you’re running Debian/Trixie headless, that’s what’s happening here.

The clean fix isn’t just redirecting output; it’s ensuring Whisper runs in an environment with a proper terminal or controlled logging. You have a few solid options:

Run Whisper inside tmux or screen so it has a real TTY:

tmux new -s whisper-session
whisper audio.mp3 --model base

Redirect Whisper’s output to a file so it never writes to stdout directly:

whisper audio.mp3 --model base > output.txt 2>&1

Run Whisper as a systemd service with proper stdout/stderr logging if you want it to run reliably in production.

The simplest and most reliable fix for your setup is using tmux or screen. It gives Whisper a clean environment to run in, keeps it headless-friendly, and prevents the I/O error entirely. This way, you’re not just patching the symptom; you’re addressing the root cause and making Whisper run well on your headless server.

[garydale]

Thanks. I suppose it makes sense since running the command then backgrounding it and exiting produces the error while staying logged in with the command running in foreground works (but is vulnerable to any system shutdown). Installed tmux, ssh'd to the server, ran the command then detached so I'll see if it works.

As an aside, I also tried manually started the virtual environment and running whisper directly but was surprised when it seemed to hang on completion. It turns out that after finishing, it doesn't return to the command prompt. The last line of output is all I see. However, ps aux | grep whisper confirmed it was no longer running, and the output file had been produced. I don't use python directly normally so is the usual behaviour?