Deploying Whisper on My NAS: Two Storage Gotchas

Background

#

Today’s project was simple: take the script I’d just written and use it to automatically transcribe audio files from Obsidian’s Whisper folder into Markdown text. The code itself was a light adaptation of the sample Python script from Whisper’s own docs — nothing complicated. Installing Whisper on the NAS, though, was a different story: mostly a string of errors caused by running out of space under root and /tmp.

Environment install issues

#

1. Create a virtual environment

cd /volume1
python3 -m venv whisper_env
source whisper_env/bin/activate

2. Point pip’s cache/temp dirs elsewhere

mkdir -p /volume1/tmp/pip_cache
XDG_CACHE_HOME=/volume1/tmp/pip_cache TMPDIR=/volume1/tmp pip3 install openai-whisper

Installing this way finally succeeded.

Runtime issue

#

When I first tried actually running it, I hit an out-of-space error again — this time because the program needs to download the Whisper model itself, and the NAS didn’t have room. Fixed it by symlinking Whisper’s cache directory over to a volume with more space:

# Make sure the target directory exists
mkdir -p /volume1/whisper_cache

# Remove the old cache directory, if present
rm -rf /root/.cache/whisper

# Symlink /root/.cache/whisper to /volume1/whisper_cache
ln -s /volume1/whisper_cache /root/.cache/whisper

After that, it ran fine — automatically transcribing voice notes from Obsidian.