Hacker News: SambhavGupta

New comment by SambhavGupta in "Local, CPU-Friendly, High-Quality TTS (Text-to-Speech) with Kokoro"

SambhavGupta — Tue, 07 Jul 2026 21:54:54 +0000

A couple months back I wrote a chrome extension that does this on any webpage, with simultaneous highlighting of the sentence being read. Skips both the container launching step and the copy pasting website contents step. Might be useful to anyone trying to use kokoro ergonomically.

https://chromewebstore.google.com/detail/local-reader-ai-on-...

Show HN: On Device AI TTS Extension

SambhavGupta — Thu, 06 Nov 2025 19:28:26 +0000

I made an on-device AI TTS extension using the WebGPU implementation of Kokoro TTS, the highest-rated open source TTS according to TTS arena. It runs model inference fully in your browser, so no server costs and it's secure.

It was pretty easy to get the TTS working since the hard part (inference in javascript/webAssembly/webGPU) was done for me. I added some quality of life features on top: click to jump to a different paragraph, alt j & k for down and up, speed control, and voices (though the default voice has the highest quality).

I made the extension mainly because I noticed how good AI TTS was getting, yet every easy way to run TTS in the browser always used the old robotic non-AI voices that are hard to listen to for more than a few minutes.

Extension is open source: https://github.com/SambhavG/tts-extension. Unfortunately it's not on Firefox as I had issues with getting WebGPU and kokoro.js on Firefox to cooperate.

Comments URL: https://news.ycombinator.com/item?id=45839228

Points: 1

# Comments: 0

New comment by SambhavGupta in "Lina Khan points to Figma IPO as vindication of M&A scrutiny"

SambhavGupta — Mon, 04 Aug 2025 02:57:29 +0000

I think it's reasonable to formulate this as two outcomes (Figma independent and Figma bought by Adobe) and each firm's preferences. From Adobe's standpoint, outcome A has huge tail risk (Figma innovates and eats Adobe's business) so paying a lot for outcome B might be ok. From Figma's POV, they aren't really helped by that tail as much as Adobe is hurt by it (either way they're rich, unless they crash out and become poor), so Figma would love to play it safe and sell the tail gamble. Adobe is paying to buy the high tail and low tail from Figma, but Figma probably doesn't care too much about selling the high tail compared to selling the low tail; meanwhile Adobe mainly wants to buy the high tail. I think it's very reasonable for Adobe to value the high tail at much more than Figma values getting rid of the low tail (Adobe is willing to bid up aggressively while Figma is willing to offer down aggressively).

Whether the government saw Adobe's willingness to overpay for Figma as a signal of Adobe's underlying incentives (as in "I acquire Figma to keep my monopoly" and not "I acquire Figma to vertically integrate and make better product for consumer") seems much harder to speculate on. I didn't explain (nor do I think there's an obvious explanation) for why the original deal was at the high price rather than the low price, but I'd imagine Figma would've been generally willing to sell for less than the price Adobe bid given Adobe was probably Figma's best customer for acquiry.

New comment by SambhavGupta in "Olympic medal counts, ranked every possible way in 3D"

SambhavGupta — Sat, 17 Aug 2024 17:19:57 +0000

It's every possible scheme which assigns a number of points for each medal. The set of all such schemes is covered by just two free parameters, for example the silver:gold ratio and the bronze:silver ratio, which are what I'm using here. Additionally assuming better medals are better, each axis goes from [1, infinity) but since medal counts are bounded, it doesn't need to be infinity in practice. The scaling is 1/x so being 1/10 from the infinity point is a 10x ratio. You could probably come up with other schemes like some weighted sum of log functions of the medal counts but I haven't seen those in practice. This is inspired by (and is essentially a 3D version of) this NYT interactive [1].

[1] https://www.nytimes.com/interactive/2024/07/29/upshot/olympi...

Olympic medal counts, ranked every possible way in 3D

SambhavGupta — Sat, 17 Aug 2024 16:47:06 +0000

Article URL: https://sambhavg.github.io/medal/

Comments URL: https://news.ycombinator.com/item?id=41275809

Points: 2

# Comments: 2

I Tried Bend

SambhavGupta — Wed, 26 Jun 2024 18:54:16 +0000

Article URL: https://sambhavg.github.io/blog/bend

Comments URL: https://news.ycombinator.com/item?id=40803252

Points: 1

# Comments: 0