Archives: Bookmarks

GitHub – nyrahealth/CrisperWhisper: Verbatim Automatic Speech Recognition with improved word-level timestamps and filler detection

Saved on September 25, 2025

CrisperWhisper is an advanced variant of OpenAI’s Whisper, designed for fast, precise, and verbatim speech recognition with accurate (crisp) word-level timestamps. Unlike the original Whisper, which tends to omit disfluencies and follows more of a intended transcription style, CrisperWhisper aims to transcribe every spoken word exactly as it is, including fillers, pauses, stutters and false starts.

100R — about

Saved on September 24, 2025

Hundred Rabbits is an artist collective that documents low-tech solutions with the hope of building a more resilient future. We live and work aboard a 10 m sailboat named Pino in remote parts of the world to learn more about how technology degrades beyond the shores of the western world.

Apples, Trees, and Quasimodes – System Stack

Saved on September 24, 2025

But there’s a subtle difference. For Engelbart, augmentation meant complexity: bootstrapping a system so wild it demanded co-evolution between user and tool. For Nelson, it meant endless layers of possibility. For Raskin, it often meant protection or constraint. Humane computing wasn’t only about empowerment… often it was about shielding users from mistakes, overload, and confusion.

AND

The humane thread survives, but only outside the center—in the tools that don’t have to answer to quarterly earnings, in projects that refuse to die just because they don’t fit the market. The Dormouse lineage isn’t gone. It just doesn’t live where the money is, because it can’t. If you want your computer to be humane in the deeper sense—not an appliance, but an instrument for thought—you have to look to the margins. That’s where it has always been, and where it still is today. If it survives, that’s where it’ll still be.

Methodology | Pew Research Center

Saved on September 23, 2025

Audio from video files was extracted and passed to an Audio Spectrogram Transformer model finetuned on the AudioSet dataset. This AST model inputs audio sequences, distinguishes speech from music, and then provides additional labels for the clip using a broad ontology of everyday sound types.
For videos where the AST model identified “speech” as the primary audio label, the full audio from the video was then passed to OpenAI’s whisper transcription model. For a balance of accuracy and fast processing time, we used the 769M-parameter “medium” version of this model. On English-language speech, Whisper performs speech recognition and transcription. On speech in languages other than English, the model also performs translation and returns English-language transcriptions.
All thumbnail images and slideshow images were passed through an optical character recognition (OCR) system using the python library EasyOCR. This OCR pass identified and extracted any text that could be read in the images.
All thumbnail images and slideshow images were also passed through moondream2, a lightweight vision language model that can perform text generation conditioned on an image. We used this model to produce short descriptions of the subject of each image.