Now working on Mac M2

[antonsalii]

I'm using a MacBook Pro M2.

I figured out that Python 3.11 should be used for this project, that's done.

Then I had several issues with the PyQt installation, but eventually I did this:
brew install qt@5

echo 'export PATH="/opt/homebrew/opt/qt@5/bin:$PATH"' >> ~/.zshrc

Then, in venv:
pip install PyQt5-sip

pip install PyQt5==5.15.10 --no-binary PyQt5 --no-deps --config-settings=--confirm-license=

Then, remove these 3 requirements from requirements.txt (since they are already satisfied):
PyQt5==5.15.10
PyQt5-Qt5==5.15.2
PyQt5-sip==12.13.0

And finally, install all other requirements using:
pip install -r requirements.txt

Then, using python run.py, the app started and I entered "English" as a language, then hit the button to save settings and restart the app, and got this in the terminal:

python run.py
Starting WhisperWriter...
No valid configuration file found. Opening settings window...
qt.qpa.fonts: Populating font family aliases took 57 ms. Replace uses of missing font family "Segoe UI" with one that exists to avoid this cost.
2025-05-02 14:11:41.393 python[93791:16317473] +[IMKClient subclass]: chose IMKClient_Modern
2025-05-02 14:11:41.393 python[93791:16317473] +[IMKInputSession subclass]: chose IMKInputSession_Modern
Traceback (most recent call last):
File "whisper-writer/src/main.py", line 106, in restart_app
self.cleanup()
File "whisper-writer/src/main.py", line 92, in cleanup
if self.key_listener:
^^^^^^^^^^^^^^^^^
AttributeError: 'WhisperWriterApp' object has no attribute 'key_listener'

Then, I run python run.py again and got this output:

python run.py
Starting WhisperWriter...
Creating local model...
config.json: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 2.31k/2.31k [00:00<00:00, 12.8MB/s]
vocabulary.txt: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 460k/460k [00:00<00:00, 5.05MB/s]
tokenizer.json: 100%|██████████████████████████████████████████████████████████████████████████████████████████████| 2.20M/2.20M [00:00<00:00, 8.95MB/s]
model.bin: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████| 145M/145M [00:05<00:00, 28.1MB/s]
[2025-05-02 14:13:56.760] [ctranslate2] [thread 16319266] [warning] The compute type inferred from the saved model is float16, but the target device or backend do not support efficient float16 computation. The model weights have been automatically converted to use the float32 compute type instead.
Local model created.
qt.qpa.fonts: Populating font family aliases took 55 ms. Replace uses of missing font family "Segoe UI" with one that exists to avoid this cost.
2025-05-02 14:13:56.940 python[93815:16319266] +[IMKClient subclass]: chose IMKClient_Modern
2025-05-02 14:13:56.940 python[93815:16319266] +[IMKInputSession subclass]: chose IMKInputSession_Modern
Traceback (most recent call last):
File "whisper-writer/src/key_listener.py", line 345, in start
self.active_backend.start()
File "whisper-writer/src/key_listener.py", line 770, in start
self.key_map = self._create_key_map()
^^^^^^^^^^^^^^^^^^^^^^
File "whisper-writer/src/key_listener.py", line 894, in _create_key_map
self.keyboard.Key.insert: KeyCode.INSERT,
^^^^^^^^^^^^^^^^^^^^^^^^
File "/.pyenv/versions/3.11.0/lib/python3.11/enum.py", line 767, in getattr
raise AttributeError(name) from None
AttributeError: insert
/.pyenv/versions/3.11.0/lib/python3.11/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
warnings.warn('resource_tracker: There appear to be %d '

How can this be resolved?

[TomFrankly]

@antonsalii I'm not the maintainer, but I've spent a lot of time trying to make Whisper-Writer work on macOS. There's a lot of work to do before it'll be compatible, and the project seems to no longer be maintained by the original dev, so I doubt we'll see macOS compatibility unless it's through a large fork.

For macOS, I use MacWhisper and am quite happy with it! It's very polished and is a one-time purchase, but if you want something completely free, check out OpenSuperWhisper: https://github.com/Starmel/OpenSuperWhisper

[antonsalii]

@TomFrankly Thanks for the info!

[rodalpho]

I recently added the new Nvidia parakeet ASR model to @TomFrankly's fork. It seems excellent, top of the ASR huggingface leaderboards.

https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2

Would post a PR but I vibed it also and it's... messy. Dependency hell with onnxruntime and onnxruntime-gpu.

[TomFrankly]

@rodalpho holy crap – one hour of audio transcribed in one second?!?

I would love to see Groq add this as a model option. My transcription with whisper-writer and MacWhisper is already near-instant since it's always very short recordings, but this would be amazing for the Pipedream workflow I use to transcribe long voice notes from my phone.

[rodalpho]

#1 on the HF ASR leaderboards; it's substantially better than the whisper-large I had been using in your excellent fork. Thanks for your work, by the way!

[smekonen]

You can resolve that issue by removing INSERT, NUM_LOCK (both of them), SCROLL_LOCK, PAUSE and PRINT_SCREEN from this map:

whisper-writer/src/key_listener.py

Lines 894 to 915 in 370333b

	self.keyboard.Key.insert: KeyCode.INSERT,
	self.keyboard.Key.delete: KeyCode.DELETE,
	self.keyboard.Key.home: KeyCode.HOME,
	self.keyboard.Key.end: KeyCode.END,
	self.keyboard.Key.page_up: KeyCode.PAGE_UP,
	self.keyboard.Key.page_down: KeyCode.PAGE_DOWN,
	self.keyboard.Key.caps_lock: KeyCode.CAPS_LOCK,
	self.keyboard.Key.num_lock: KeyCode.NUM_LOCK,
	self.keyboard.Key.scroll_lock: KeyCode.SCROLL_LOCK,
	self.keyboard.Key.pause: KeyCode.PAUSE,
	self.keyboard.Key.print_screen: KeyCode.PRINT_SCREEN,
	# Arrow keys
	self.keyboard.Key.up: KeyCode.UP,
	self.keyboard.Key.down: KeyCode.DOWN,
	self.keyboard.Key.left: KeyCode.LEFT,
	self.keyboard.Key.right: KeyCode.RIGHT,
	# Numpad keys
	self.keyboard.Key.num_lock: KeyCode.NUM_LOCK,
	self.keyboard.KeyCode.from_vk(96): KeyCode.NUMPAD_0,
	self.keyboard.KeyCode.from_vk(97): KeyCode.NUMPAD_1,

I also had to remove the line below for the start button listener and just start it directly with self.key_listener.start() because of this issue: moses-palmer/pynput#511

whisper-writer/src/main.py

Line 59 in 370333b

self.main_window.startListening.connect(self.key_listener.start)

Those workarounds (and changing the PyQt versions) were enough for basic Mac support on M1, at least for the features I've tried