Abhinav Prakash @NeurlAP

Senior Software Engineer(Gen AI) @PocketFM_App | Ex Backend @mydukaanapp | Former Lead Google @DscGita |Co-Organiser @PyDataBbsr | Founded xieken-tech neurlap.in Bangalore Joined January 2018

Tweets

6K
Followers

1K
Following

4K
Likes

13K

prateek dixit @prateek1190

24 hours ago

I've been thinking about why there's such a large gap between production grade voices for entertainment and what existing TTS models actually deliver. The reason is simple: most of these models are optimised for low latency use cases. Enterprise, call routing, voice assistants. Speed is the metric that matters. But entertainment doesn't need that at all. At @PocketFM_App , we're building high quality audio series and serialized fiction. Nobody here loses sleep over 200ms response times. What keeps us up is whether a character's whispered confession sounds different from their battle cry, or whether both get the same flat, lifeless delivery that makes a listener quietly close the app. Entertainment-grade TTS is genuinely unsolved. The research community knows it. Our team is building the solution. Congratulations to the @PocketFM_App research team on getting this work accepted at ICML. Quietly shipping research while building a product is hard, and this one matters. The approach is elegant: improve emotional delivery without touching the synthesis model at all. The deeper insight: you don't need a bigger model to get expressiveness. You need the right reference. And the right reference, at scale, is a data problem. The amount of golden data @PocketFM_App has, high quality, narrative, expressive, across languages, that's going to be the defining factor in solving this. Not just for Hindi. For every language we touch. We're taking this further, not just prosody and emotion in speech, but the full production stack. BGM, SFX, mastering, emotional continuity across an episode. The kind of audio that makes someone forget they're not listening to a studio production. Nobody's really solved entertainment-grade TTS + audio mastering end-to-end yet. That's exactly why we're working on it. If this problem excites you and you want to work on it, DM me or reach out at [email protected].

0 1 3 84 0

View Details

Abhinav Prakash @NeurlAP

2 days ago

😭 I'm once again asking for good broker contact for Harlur, sarjapur, near bellandur"

Striver | Building takeUforward @striver_79

2 days ago

Rounak Adhikary @Rounacc

2 weeks ago

I waited 4 years for this day. My mom told me once “closed mouths don’t get fed” I listened to her

154 278 5K 184K 868

View Details

Abhinav Prakash @NeurlAP

2 weeks ago

@arcinston Via which service

0 0 0 87 0

View Details

Abhinav Prakash @NeurlAP

2 weeks ago

@iNishant Actions down 😂

0 0 1 50 0

View Details

Abhinav Prakash @NeurlAP

2 weeks ago

@0xastro98 @GruhamBot Can you add me

0 0 0 83 0

View Details

Rohan Nayak @RohanNayak2

3 weeks ago

Big moment for storytelling AI research at @PocketFM_App Two papers accepted at International Conference on Machine Learning on problems we care deeply about: • Long-horizon narrative intelligence • Cultural adaptation in storytelling Most AI systems today optimize for prompts. Real entertainment needs memory, pacing, emotional consistency, and cultural nuance across hundreds of episodes. Proud to see our research team pushing the frontier on narrative world models and culturally adaptive generation. More to come.