apoorv

@Devanshpawan1 @smallest_AI @StalwartCoder @harshitajain561 @itsnotraunaq @indurajaditya @himanshustwts @StalwartCoder why do u look so dead bro

Say hi to the growth team @smallest_AI

@smallest_AI team is cooking 🫡🔥 TIL they just dropped an open-source Wispr Flow alternative. Makes total sense. Their voice models are already some of the best. OSS = way less stress about where your data goes. github.com/smallest-inc/M… @kamath_sutra & team ftw 🚀

@devxMani Will still choke

@apoorvcodes they r the best in the country rn I don't think so

🚨 Bengaluru and SF-based @smallest_AI has launched their SoTA TTS model, Lightning V3.

I cloned Peter Griffin's voice and made him do a code review. 10 seconds of audio. One API call. ~90ms latency. This is 100% AI-generated. Code attached.

We identified where pronunciation lives using acoustic causal tracing, then applied a closed-form, one-shot weight update which enforced null-space constraints ensuring zero interference with existing behavior. No retraining, no extra parameters, no degradation in prosody or speaker identity. Editing generative models like databases precise, local, and efficient.

Where does pronunciation live in a large language model(LLM) based text-to-speech(TTS) system, and how can we surgically modify it for specific texts while preserving all other model behavior? To answer this very question, we introduced SonoEdit at @CPALconf yesterday. Our core hypothesis is that pronunciation errors aren’t global but they live in localized internal representations. If you find them precisely, you can fix them precisely.

@nityanandmathur aree aap sir, itne bade words ke liye shukriya

@apoorvcodes Legend!

Give it a read guys!!!

@Hi_Mrinal @sumir_30 dekhra hai bhai kya kr rha hai ye

Some conversations which I think should be publicly available 1/n > I don't want to sound dumb but like, even just organizing code into folders counts as architecture right ?? >> That is what you call modularity, when you add folders you are just trying to separate concerns which is also a good habit. But there's the thing it's only a small piece of architecture. Modularity is more of a coding practice thing than a full architecture decision. What it does help with is debugging, testing, and keeping your code easy to maintain.

@Hi_Mrinal kaam krle bhai

@apoorvcodes no

@nityanandmathur @A_a_yush @CPALconf @smallest_AI 🫡🫡🫡🫡

Me and @A_a_yush are at @CPALconf, and will be presenting SonoEdit(arxiv.org/abs/2601.17086) in the evening. Come talk with us about speech-native intelligence, speech to speech and text to speech. @smallest_AI

crackedddddd

realizing ChatGPT + Claude can build your whole app but your voice agent still sounds like a robot because TTS was optimized for reading not conversation. Smallest AI just changed that MOS 3.89 beats OpenAI, ElevenLabs, Cartesia and yeah… you can hear the difference instantly

Can't wait to complete and put out what I'm working on..

It's amazing how Lightning V3 just dominates every single TTS out there, across every parameter that matters. And this is not even our best release this month!

Super proud of the team @nityanandmathur @wasim_madha 💪💪

Keeps impressing - honestly, I can't believe how human it sounds!

@nityanandmathur aapse aur @AkshatMandloi10 hi seekhte hai sab sirji

@apoorvcodes Whole smallest team is cracked, sir! 🤗

The voice cloning is justtt out of the world!!! @nityanandmathur is cracked aff