Read this book https://ccrma.stanford.edu/~jos/pasp/. It's a hefty tome but kind of the bible of the various approaches to audio signal processing and synthesis.
There's no one method to rule them all, they all have tradeoffs. So there is no "ray tracing" of audio (well actually, it's ray tracing, but for a lot of reasons it's prohibitive to actually do that in a way that sounds good).
> What is this field called?
Audio synthesis. Popular journals with papers covering the topic are DAFX (digital audio effects) and jAES (journal of the audio engineering society). To a lesser extent, the IEEE transactions on audio signal processing.
> I am assuming it has something to do with the physics field of acoustics?
You can find plenty of sources in acoustics journals/text books - but this is like comparing the needs of mechanical and electrical engineers to game engine designers. At a surface level there is crossover, and cross pollination of techniques/tools, but the needs are fundamentally different.
If you want to learn how to understand the performance of the whole system I can recommend Brendan Gregg's Systems Performance: Enterprise and the Cloud (https://www.brendangregg.com/blog/2020-07-15/systems-perform...). It is a good book that teaches a lot of basics and techniques and gives a good understanding of the impact different system components can have on performance.
Been working on this as a new way to find, explore, and view old historical maps and aerials of my area. Still heavily in development but have been surprised by some early traffic stats (1-2K organic uniques / mo and growing 200%+ m/m right now)
Hoping to add in more advanced map tooling within the next week or 2, including new basemap options, 3d terrain view, and then a proper search box which I've been pushing off for far too long
I used to be a full time dev / R&D engineer. Now I basically do the same thing on YouTube (youtube.com/stuffmadehere). The difference now is my R&D-ing is directed at early stage prototypes that I think are interesting / instructive, rather than what is best for an actual business, useful, or profitable.
Youtube is interesting because theres a constant source of numeric feedback on how you are doing (views / subscribers / watch time). Seeing these numbers change based on what you do can be incredibly addicting and it's very easy to accidentally connect your personal happiness to those numbers. This is great if they are going up, but if they aren't.... yeah. It's also easy to get into a situation where you lean into "what works" over and over until you find yourself doing stuff that you don't enjoy.
My advice would be to find a way to keep the numbers at arms length and focus on doing stuff that you enjoy. You definitely need the feedback of stats / comments / etc to get better, but you don't wan to check it 10 times a day. Personally when I launch a video I will check a few times to ensure I didn't screw up anything major, see if there is any useful feedback in comments, then I will check maybe the stats every week or two.
Judging by the URL, this book was used for CMU's 15-151 / 21-128, which is a first-semester course for CS and math undergrads. Nowadays, the course uses [0].
LinAlg was the only maths course I needed in my interdisciplinary study program. I had struggled to grasp maths in high school, but these lectures really made it click for me and I passed my university's class with a B+.
See also Inc's writeup: "This Hollywood Director's 2+2 Rule Will Make You Instantly More Persuasive and Charming. This incredibly simple rule for better storytelling is also incredibly powerful."
"We would call this the unifying theory of two plus two. Make the audience put things together. Don't give them four; give them two plus two. The elements you provide in the order you placed them in is crucial to whether you succeed at engaging the audience."
AWS GPUs (or ones offered by any of the big cloud providers for that matter) are not really all that cost effective. They also have to use datacenter GPUs, which are a lot more expensive than consumer ones.
For cost effectiveness, one route (mostly in the ML community) is to rent people's machines via services like vast.ai, who can get around NVIDIA preventing use of consumer cards in data centers by making it so they're only selling/buying time on people's computers.
These end up being a lot more affordable (IIRC ~$20/day for a 2x 3090 system) but in exchange have the unreliability that comes with renting time on someone else's computer, like tasks randomly getting killed (which is pretty frustrating to wake up to, as you get charged for time rather than usage), some people having poor upload/download speeds or other gpu performance issues.
If you suffer from this kind of perfectionism and find it affects your productivity (i.e. in the form of procrastination) I highly recommend reading The Now Habit by Neil Fiore. It's helped me to not only be more productive, but also to feel less guilty and more confident about myself no matter how much I actually get done.
For anyone interested, the original design doc for QUIC from 2013 [0]. Really good writeup, both in terms of engineering spec / architectural design. I recommend reading through if you have the time.
Here's a video lecture from the MIT Professor (Dennis Whyte) who was leading the research group that provided some of the key designs for the SPARC reactor. As the NYT article explains, that research has been spun out into a startup that raised $200M.
The key breakthrough is the advancement of REBCO tape superconductors which allow you to (1) generate record breaking magnetic field strengths (2) easily disassemble the super conducting loop for fast repairs / refuels / more modular design.
It's a long talk, but it's extremely fascinating. Basically everything becomes much easier once you can increase the magnetic field strength. This talk is fairly accessible to even relative laypeople who have a vague understanding of E&M physics.
I can make a blog post later, but at a high level:
A rust TTS server hosts two models: a mel inference model and a mel inversion model. The ones I'm using are glow-tts and melgan. They fit together back to back in a pipeline.
I chose these models not for their fidelity, but for their performance. They're 10x faster at inference than Tacotron 2. If you want something that sounds amazing, you're better off with a denser set of networks, like Tacotron 2 + WaveGlow. You should use these for achieving superior offline results for multimedia purposes.
Instead of using graphemes, I'm using ARPABET phonemes, and I get these from a lookup table called "CMUdict" from Carnegie Mellon. In the future I'll supplement this with a model that predicts phonemes for missing entries.
Each TTS server only hosts one or two voices due to memory constraints. These models are huge. This fleet is scaled horizontally. A proxy server sits in front and decodes the request and directs it to the appropriate backend based on a ConfigMap that associates a service with the underlying model. Kubernetes is used to wire all of this up.
IIRC, some company wanted to use SQLite on an airplane, so they paid the devs enough to bring the test suite up FAA standards. IIRC, they have code coverage of every machine instruction.
(copy-pasted with modifications from my comment on a similar thread posted earlier)
I used to have an elaborate system, but I converged on a simple solution: I stash everything in a single Google Docs document.
I made the conscious decision to optimize for ease of use, so that the friction/effort to write something down is minimized.
At the same time, I also made a decision to not to adopt any organization system -- anything that increases the friction of use is eschewed. Search was all I needed. I've been using this system for the past 5 years or so, and it's been very productive.
To extract ideas from it, I routinely re-read stuff (it's in log format, so it's very easy to read) and use the Fieldstone approach (Weinberg)[1] to coalesce similar and interesting thoughts and rewrite into larger thoughts. I've gotten a lot of actionable ideas this way (that I actually go on to execute on).
So it's a system optimized for postprocessing rather than pre-processing. (The Fieldstone approach is a method from writing, rather than knowledge management.) I find that pre-processing systems are unsustainable over the long term unless you're exceptionally disciplined (also, it's hard to know how to structure knowledge until you've processed it -- most interesting ideas are a garbled mess when first encountered). I try to build systems that don't rely on sustained human discipline or the necessity of shoehorning into known organization units.
I'm shocked at how much negativity there is here for Inkscape - it's long been a go-to example for how great open source software can be, in my mind.
These days, I often use the (non-free, $26) Vexlio [1] software since it works better for my use-case (diagrams for use in Latex documents). But Inkscape is clearly the more fully-featured product for most users.
If you're interested in the history of the Titan missile program and US nuclear program in general I recommend "Command and Control: Nuclear Weapons, the Damascus Accident, and the Illusion of Safety".
No, I drew Inconsolata myself, as well as inventing spiral-based tools for designing fonts, as an alternative to Béziers. I started work on it far in advance of working on the "Google Font API" as it was called at launch, so it was very natural to include it, along with other great fonts by other talented designers.
I ran a website for youngsters several years ago.
One of the duties to maintain it was to moderate discussion boards.
Some kids were difficult to manage and would not accept to be banned (via email/IP/or whatever solution) and would keep recreating profiles.
Ultimately I dealt with those ppl by “greylisting” them. Added a sleep() prior each page rendering of 5 to 25 secs (actually it was more sophisticated and would stream chunks over TCP so the slowness feeling was even more real).
Worked like a charm. Few days after the recalcitrant would no longer come on the website.
I called this “moderation by degradation of user experience”, and was pretty effective like the solution described in your post.
Think about page load if you need to restrain visits.
