Flatpak Speech Note crashes when using Faster-Whisper: ctranslate2 cannot enable executable stack (Per

[0vermind]

Hi,
On openSUSE Kalpa (immutable) on an Intel laptop, the Flathub Flatpak of Speech Note crashes as soon as I try to use the microphone / start STT with Faster-Whisper.

The terminal output shows an ImportError from ctranslate2:

ImportError: libctranslate2.so.4: cannot enable executable stack as shared object requires: Permission denied

Then Speech Note fails to create the faster-whisper model and aborts.

Steps to reproduce:

1. Install Speech Note from Flathub (net.mkiol.SpeechNote).
2. Open the app, download Whisper models (Faster-Whisper).
3. Click the microphone / start transcription (Faster-Whisper backend).
4. App crashes.

Expected:
Microphone starts transcription.

Actual:
Crash during faster-whisper model creation (ctranslate2 import fails because executable stack is blocked).

Environment:

• OS: openSUSE Kalpa (immutable) + KDE Plasma
• Hardware: Lenovo ThinkBook 14 IIL (Type 20SL), Intel CPU
• Flatpak app: net.mkiol.SpeechNote (Flathub), version 4.8.3
• Runtime: (please fill from flatpak info net.mkiol.SpeechNote)

Full log:
flatpak run net.mkiol.SpeechNote
[E] 09:56:27.457514732.457 0x7f17b57fb680 operator():155 - py error: ImportError: libctranslate2.so.4: cannot enable executable stack as shared object requires: Permission denied

At:
/app/extensions/cpu/lib/python3.12/site-packages/ctranslate2/init.py(53):
(488): _call_with_frames_removed
(999): exec_module
(950): _load_unlocked
(1333): _find_and_load_unlocked
(1360): _find_and_load
/app/lib/python3.12/site-packages/faster_whisper/transcribe.py(10):
(488): _call_with_frames_removed
(999): exec_module
(950): _load_unlocked
(1334): _find_and_load_unlocked
(1360): _find_and_load
/app/lib/python3.12/site-packages/faster_whisper/init.py(2):
(488): _call_with_frames_removed
(999): exec_module
(950): _load_unlocked
(1333): _find_and_load_unlocked
(1360): _find_and_load
[E] 09:56:27.457591087.457 0x7f17199ff680 create_model:179 - failed to create fasterwhisper model
[E] 09:56:27.457626706.457 0x7f17199ff680 process:339 - stt processing error: failed to create fasterwhisper model
/usr/include/c++/14.2.0/optional:475: constexpr _Tp& std::_Optional_base_impl<_Tp, _Dp>::_M_get() [with _Tp = pybind11::object; _Dp = std::_Optional_base<pybind11::object, false, false>]: Assertion 'this->_M_is_engaged()' failed.

Notes:
This looks similar to the known ctranslate2 “cannot enable executable stack” problem on hardened environments / newer glibc behavior.
A possible fix might be rebuilding/patching the shipped ctranslate2 library to not require execstack (e.g. proper .note.GNU-stack / -z noexecstack, or clearing the execstack flag during packaging).

[mkiol]

Hi. Sorry for late reply and thank you for reporting this problem.

Very interesting. CTranslate is built from sources, so rebuild (or re-link) is perfectly possible.

Fix: 58b989c

I also pushed a new flathub test build. Once this is complete, I will ask you to re-test it again.

[mkiol]

Here is a test version that includes the fix:

flatpak install --user https://dl.flathub.org/build-repo/251426/net.mkiol.SpeechNote.flatpakref