getting blurt to use the CUDA whisper.cpp version in Fedora Silverblue

an immutable OS means never being able to install the CUDA toolkit directly In order to get the blurt gnome desktop integration to use the whisper.cpp CUDA model in Fedora Silverblue (or any immutable OS), you need to do a couple of things: main-cuda is the key Assuming you have whisper.cpp and blurt installed and…

distil-whisper is the tipping point

astounding accuracy of OpenAI's whisper is retained with massive speed improvements in distil-whisper transcribing the excellent 40 minute Prosecuting Donald Trump podcast in 45 seconds If you have ever suffered through voice-to-speech software from the 1990s with its awkward setup, convoluted training, and extremely poor accuracy, you will be astounded by the latest automatic speech…

Testing out LLMs with a Mandelbrot plot using numba and CUDA

I don't think LLMs understand CUDA I am exploring numba and CUDA. What better way to examine code than to ask an LLM to generate samples. It works for some code examples, but it is painfully obvious that at best LLMs are a copy/paste function (with lacuna) and at worst simply don't understand CUDA at…

How to transcribe hundreds of audio files into text using OpenAI Whisper

As I mentioned earlier, OpenAI's Whisper transcription model is the current state of the art, with stunning results. Whisper is so good, in fact, that I felt it worthwhile to transcribe hundreds of old audio files into text. Rather than doing this manually, I set up a workflow, which you can see in the enclosed…

I am stunned by OpenAI Whisper’s voice to text results

The following blog entry was essentially transcribed (with manual editing where I was unhappy with what I said). OpenAI Whisper using the medium model (unedited with errors in red): The new OpenAI Whisper Voice to Transcription Service is now by far the best transcription service available. It is open source and can very easily be…