Please add Ai Transcription also on Mac (OpenVINO™ Whisper Transcription)

bittenbyajerk · January 23, 2024, 9:49pm

I was looking for similar plugin for quite some time. This would make transcribing and exporting various sequences breeze… + much easier orientation in the timeline or organizing in exported files too…

Would be difficult to support also Mac aside from Win and Linux?

github.com

intel/openvino-plugins-ai-audacity/blob/main/doc/feature_doc/whisper_transcription/README.md

# OpenVINO™ Whisper Transcription :microphone:

This feature performs speech transcription or translation for a mono or stereo track, producing a label track.

It can be found under the **Analyze** menu:
![](menu.png)

## Description of properties  
See below for a description of the properties that can be set for this analyzer:  
![](properties.png)  
1. **OpenVINO Inference Device**: The OpenVINO™ device that will be used to run the whisper encode model. This should default to something reasonable.
2. **Whisper Model**: The whisper model to use. Right now, we are only supporting *base* model, but support for others may come in the future.
3. **Mode**: Select between *transcribe* and *translate*. Set to *translate* if you want the output to be in english, regardless of the source language. Set to *transcribe* if you want the output to be in the same language as the source audio.
4. **Source Language**: Gives the ability to explicitly set the source language. The default is *auto*, which will perform auto-detection.


After clicking *Apply*, you'll see this dialog window pop up:  
![](loading.png)

At this stage, the whisper AI model is getting loaded to the chosen device (e.g. CPU, GPU, NPU, etc.). This usually takes 10 to 30 seconds if it's the first time running with this device after installing these plugins since it needs to compile the model specifically for the device you've chosen. These *compiled* model will be cached on disk though -- so it should run much faster the next time that it is loaded.

This file has been truncated. show original

RobBW · January 24, 2024, 3:32am

Would this help?
Hear. macOS speech recognition via the command line
**hear** is a command line interface for the built-in speech recognition capabilities in macOS. This makes it possible to transcribe spoken audio input or entire audio files.

bittenbyajerk · January 25, 2024, 7:18am

Thanks for the tip. Unfortunately transcribing and making a label region so I can export each sentence separately with its content in the file name would help me the most.

There are other algorithms for speech recognition on Mac but none are integrated with Audacity.

I’ve found whisper ai, supporting more languages and can transcribe files locally.
Theoretically there is a way to make a python script or something with Whisper and Audacity. I just don’t have enough experience to make it :)) maybe chat GPT or something like that might help building it.

adamjohansen31 · January 25, 2025, 10:29am

If you want to get in touch with someone experienced in Python, they might help create a script linking Whisper AI and Audacity—it sounds doable!