You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Transcription fails when the disk runs out of space #75
When free disk space is low, transcription can fail outright. Before transcribing, AutoDoc converts the meeting audio to a temporary audio file. For long recordings, or a recording whose length is reported far longer than the meeting really was, that temporary file can grow to several GB and fill the disk. The conversion then fails, the meeting gets no transcript, and the partial temporary file is left behind, so the disk stays full.
The same thing can happen when the speech model downloads and installs: the install fails with "No space left on device".
What still works
The recording itself is still saved. Meetings transcribe normally when there is enough free space. A speech model that has not downloaded yet is #67.
How to reproduce
Fill the disk until only a little free space is left.
Record a long meeting and stop it, or start a speech model install.
Expected: AutoDoc warns before starting that there is not enough space and says how much to free, instead of failing partway through and leaving a large temporary file behind.
Environment
AutoDoc version: 1.3.0 and 1.3.1
OS: macOS and Windows
CPU / RAM / GPU: not specific to one machine
Error
Transcription fails during audio conversion with no clear disk-space message. The speech model install fails with "No space left on device".
Workaround
Free up several GB of disk space, then click Retry on the meeting, or restart the speech model install. There is no in-app way yet to clear a leftover temporary file from a failed conversion.
Done when
Before converting a meeting's audio, AutoDoc checks free space and, if it is too low, warns right away with how much space to free.
After a failed conversion, the temporary audio file is cleaned up so the disk is not left full.
Free space is checked before the speech model is installed, with a clear disk-space message instead of a generic failure.
A recording length that is far longer than the meeting actually ran is caught before conversion, instead of producing a huge temporary file.
Activity