Hacker News
Productive Procrastination
Productive Procrastination
Productive Procrastination
I gave every train in New York an instrument
Understanding the Kalman filter with a simple radar example
Understanding the Kalman filter with a simple radar example
Understanding the Kalman filter with a simple radar example
Show HN: I pipe free sports streams into Jellyfin – no ads, just HLS
Show HN: I pipe free sports streams into Jellyfin – no ads, just HLS
Understanding the Kalman filter with a simple radar example
How Complex is my Code?
X / Twitter
Software Architecture Guide. By Martin Fowler.
As a designer, I truly dislike moodboards, specifically moodboards that are just full of end work from others, and here's why. I've been teaching design for 8+ years and mood boarding is traditionally one of the first exercises I give to students. Yet the reason why mood boards exist is often misunderstood. Mood boards shouldn't be just a collage of final work you love. That's a reference board, not a mood board. Their purpose is to capture a feeling, a vibe, a world, not final outputs. So the best materials to put in a mood board are: textures, colors, illustrations, archival photos, typography in use, photography, environments. The right combination of all this leads to a tonal quality. Reference boards are different and have their place. But by definition, they make you and your clients think inside the box. They communicate: I want to make something that looks like X and Y. That keeps you from creating original work. A mood board guides inspiration and conversation early in the process. It says: I want to make something that feels like these textures, these colors, these spaces. So when making mood boards with students, they must put 0 final designed work in their mood boards, and this exercise always leads to more interesting and original results. Mood boards are a tool for discovery. Reference boards are a tool for direction. Both have value, but they're not the same thing. If you're using the terms interchangeably, you're likely doing one of them wrong.
It's 1:30 am and I've nothing better to do, so here we go: stereo vs. mono: yes, if you have stereo sound, the ai speech is usually on the left channel, and on the right is the original audio. To mute AI speech, mute left channel. noise: most news shows are extremely aggressive with noise reduction and audio compression. This will cause audible artefacts in the generated ai speech. Older footage/lower quality footage tends to produce less artefacts. speed: the ai speech generator is roughly limited to 10-15 seconds of audio per second in realtime. This is why live is not feasible. On 4090 it's roughly 30s/s. translation delay: it takes around 10 seconds to recognize, translate and then generate the audio. why is speech sometimes out of sync? The transcription picks up full sentences to translate. The speech generator reads the sentence when it's done. This means that the AI voices try to cram all the audio in the original time-frame of the sentence, leading to sometimes faster or slower speech. future dev: currently hard-coded to translate to English. Future plan is to be able to configure the input and output language. Also planning to test it on original language sources
sam3.cpp - Meta's SAM 3 in pure C++ with @ggerganov's ggml - Supports SAM 3.1, 3, 2.1, 2 and Edge variants - Runs on CPU or GPU - GPU + CPU pipeline for better performance. github.com/YavorGIvanov/sam.cpp
recommended reading.
[Article post - text in article card]