Noah

Building Silestis & SonWeave. Writing about speech recognition, local compute, and turning audio/video into useful knowledge.

[Noah @noahdev27]

Hey @X
Silestis is now on iPhone and iPad. Open a video in Safari, turn what was said into searchable text, and ask if it covers what you need before watching it all.
https://t.co/PRaYeUZJ2q

[Noah @noahdev27]

@codebyjorge Happy to connect

[Noah @noahdev27]

@soy_muse I’m building Silestis so you can see whether a long web video covers what you need before you watch it. You can search what was said instead of guessing from the title or thumbnail.

[Noah @noahdev27]

An offline transcription feature still has to run on someone's actual phone. Battery use, free space, and support for older devices are part of whether it's useful.

[Noah @noahdev27]

Chapters help until the answer is buried halfway through a long lecture. Without a timestamp on the search result, you're scrubbing through the video again.

[Noah @noahdev27]

A recording with two people talking at once can produce a transcript that reads cleanly and still gets the speaker wrong. That matters when you're quoting someone.

[Noah @noahdev27]

If an interview transcript gets a name wrong, searching for the right name can return nothing. Fixing the transcript should fix search too.

[Noah @noahdev27]

@NVIDIAGeForce #RTXON

[NVIDIA GeForce @NVIDIAGeForce]

We're giving away a CONTROL Resonant custom wrapped NVIDIA GeForce RTX 5080 to celebrate its official release with #RTXON.

To enter:
🟢 Share this post
🟢 Comment #RTXON

T&Cs: https://t.co/Mbf7w3j1q3

[Noah @noahdev27]

@asmehra95 That right Alt shortcut sounds handy. If it gets a name wrong, can you fix it at the cursor without starting over?

[Noah @noahdev27]

@Sim_Onchain @Ozacle23 Would jumping from a transcript line back to the clip help, or is it the whole conversation that makes it stick?

[Noah @noahdev27]

@gkemonapp Both meanings of transcript are valid, which makes this easy to miss. Did you change the English source string to 'transcript of the video' or add translator context for those 63 uses?

[Noah @noahdev27]

@ArkDavey @SeanTrende Has a quote ever looked right in the transcript but meant something different once you heard the audio around it?

[Noah @noahdev27]

I keep coming back to a simple test for AI products: what still works when the connection drops? Not every task belongs on a device, but that question changes which features you build locally.

[Noah @noahdev27]

One thing building speech tools has taught me: fast processing is only part of the time someone spends. Importing a file, checking the transcript, fixing names, and exporting it can take longer. I want to measure the whole workflow.

[Noah @noahdev27]

A voice memo is easy to record and easy to forget. I built SonWeave to make the recordings already on your phone searchable. Import one, find the part you need, and go back to the audio. The US price is $4.99 once. https://t.co/Me5lrEy3nO

[Noah @noahdev27]

For a long podcast, a short summary often skips the detail I actually came for. I would rather search the full transcript, find the relevant passage, and listen to that moment myself.

[Noah @noahdev27]

@LearnCopilot How do you verify that an action item was actually agreed on, rather than just discussed?

[Noah @noahdev27]

@HuggingModels Interesting model. Have you seen WER and real-time-factor benchmarks on CPU vs Apple Silicon, especially for longer recordings? The size, speed, and accuracy tradeoff would be useful to compare.

[Noah @noahdev27]

@sharifelkomi @alymoursy @WisprFlow Losing a transcript would outweigh a small accuracy gain for me too. Do you remember if it disappeared during recording, processing, or after it had appeared to save? I build transcription tools and am trying to understand where reliability breaks.