shiv.

Building AI for music generation & understanding.
Research • Open Source • Agentic AI

[shiv. @sonic_inference]

Hey, I am Shiv

Let’s not go for a formal introduction. I was always into music since my class 4, and here I am , certified Indian classical musician (8+ years with 3+ years on stage experience ). But maybe I was naive or dumb, never once it came to my mind that I should pursue Music as a career, and I just saw my cousin elder brother and choose Engineering, so it was nothing dramatic for me like any passion , goal , achievement for me. And I am into engineering, I want something to do in life that keeps me connected deeply to my passion and also that I will enjoy my job much more .

Now I am CSE- AIML graduate, yes I have done good-decent projects , some hackathons( both participated and organised ), in my third year I fully focused on projects (on which I fully focused ). Now I have completed DSA too, but I don’t wanna sit for on campus placements right now (unless I confirm that I have freedom to apply only to selected companies , coz I just don’t want to get selected somewhere and call it a job)

Coming to this account, it will be purely a technical/professional portfolio for myself, where I would be sharing my updates of something I love working on : AI+MUSIC

Yes, I just want to work in this field because I have so many delusions into this. I have also experienced some limitations to world class music models like Lyria and so many others, hope I become capable of contributing

And this is all , any suggestions drop down below!

[shiv. @sonic_inference]

@aaryanngoyal Keep up

[shiv. @sonic_inference]

Do you think this approach is especially valuable for curiosity research AI + Music for me, where much of the work is about connecting concepts across fields rather than solving a well-defined engineering problem? my best ideas usually have appeared while trying to connect insanely different and unrelated concepts.

[shiv. @sonic_inference]

@Col_ASY even i started my 4th year , planning to do the same, hence started this account a week ago , hope everything works out well for everyone

[shiv. @sonic_inference]

Revising Binary Search Trees today

[shiv. @sonic_inference]

@xaemio I have the upper right one

[shiv. @sonic_inference]

@mohinitwt Thank you 🤞

[shiv. @sonic_inference]

music intelligence, 5

started with the nature of sound , the resources i will be referring :
Open Yale Courses
Lecture 18 - Simple Harmonic Motion (cont.) and Introduction to Waves
(https://t.co/vuPfAeztpj)

[shiv. @sonic_inference]

@xaemio caffeine has become like my study partner

[shiv. @sonic_inference]

https://t.co/njSRVR4fln

[shiv. @sonic_inference]

@Pseudo_Sid26 Keep growing brother

[shiv. @sonic_inference]

@xaemio yep, hoping for best , you too best wishes

[shiv. @sonic_inference]

@sankitdev @xaemio yep brother , we just have to believe in ourselves

[shiv. @sonic_inference]

@freshlimesofa @xaemio damn bro, nice to see you again

[shiv. @sonic_inference]

@neuroquark Monster ?

[shiv. @sonic_inference]

@freshlimesofa @xaemio hey bro , are you the one with that greyish face pfp ?

[shiv. @sonic_inference]

@xaemio now i have completed DSA almost and yet , dying

[shiv. @sonic_inference]

facing the same problem here , idk whether to sit on-campus or go for off, ofc it's not a skill issue, the problem is , we are not getting even interview and this is the dm i sent to someone 2 days ago :

Hello sir, I am Shiv, read your post about not doing DSA ,
Tbh, I am stuck with the same problem, I literally have to force myself to even solve one question of DSA, like no interest, nothing. I have left and restarted many times , but never finished. Instead I love doing my AI+MUSIC thing, I have started getting deep into this, like research papers, blogs, articles and all

But here’s the problem, I am not able to decide whether i can afford not doing DSA, coz i just started my 4th year and everyone has started placement prep through dsa. When i talked about it to my parents , they said they wont even worry if i get not placed even after a year of graduation , but never ever get into wrong job where you will not see personal growth. Can you please guide me through this sir

[shiv. @sonic_inference]

https://t.co/iapJsROQL3

[shiv. @sonic_inference]

music intelligence , 4

this weekend , i published my first blog post breaking down the new Magenta Real-Time 2 (MRT2) paper.

while reading this the biggest question for me was How do you teach a computer to imitate something deeply expressive , like a specific playing style of a musician or a player ?some options were symbolic data MIDI , which is compact/great for structure but loses that FEEL , or a raw audio which captures the exact texture but is a big expensive pile of numbers to process

now that creates a new problem : Latency
MRT1 had a 3sec delay , which breaks real-time interaction and MRT2 drops this to just 200ms by spitting the architecture into a temporal transformer (using sliding window attention to only look at recent context ) and a depth transformer. the below visualisation clearly shows the exact flow from the encoder to the scheduler , and how tokens move between the swa cpu cache and the paged kv gpu cache in real time

i have attached the links in the post below

[shiv. @sonic_inference]

@mohinitwt @solana hey , you use any app for rec ? and subtitles one