Get real-time transcriptions for anything playing audio on a Mac. Targeted for English output.
Please do the following sections in order shown.
Install Blackhole 2ch For Mac: https://existential.audio/blackhole/
- This is needed to send audio to whisper for transcription
- Click the
+icon in the bottom left and choose the option with 'Multi-Ouput Device' - Configure the Multi-Output Device to look this (enable drift correction for all devices to use besides Blackhole):
- Go to
Blackhole 2chin Audio MIDI, and make sure Output format is 48,000 Hz and Primary value is 1.0:
- Now every time you want audio to go to both speakers and Blackhole, just change the Sound output to Multi-Output Device:
Warning
In Multi-Ouput Device output, the volume will NOT be able to be changed. You must change the volume of the speaker first before switching to Multi-Ouput Device.
Install Xcode from App Store and accept Agreement
- Make sure you have conda installed (https://docs.conda.io/projects/conda/en/stable/user-guide/install/index.html)
- Run
conda env create -f transcription.yml - Then activate using
conda activate whispermac
Now run bash setup.sh, and then wait for the set up to be complete!
Finally, run python main.py to start subtitle transcription. Note, it can take 5-10 seconds to start up on first run. Subsequent runs will be instant.
- Run
conda activate whispermac - Run
python main.py
cdintowhisper.cpp- Build binary with
cmake --build build -j --config Release - Run
./build/bin/whisper-stream -m ./models/ggml-large-v3-turbo.bin -t 6 --step 1000 --length 5000 --keep 500- This should use the microphone by default. If want to switch to Blackhole, look at the list of devices shown from the capture devices list and run the command with
-c <number corresponding to Blackhole>
- This should use the microphone by default. If want to switch to Blackhole, look at the list of devices shown from the capture devices list and run the command with
After pip install, run conda env export > transcription.yml
Run black *.py