It's not striking because a person who wears an eye patch and has a tendency towards dark clothing is stastically more likely to be involved in an accident where seeing and being seen are important.
Hey, thanks for this. Been trying it out and it's very fast but seems to hear more speakers than are in the audio. I didn't see a way to tweak speaker similarity settings or merge speakers in some way. Any advice?
Yeah unfortunately, since the diarization is acoustic features based, it really does require high recorded voice fidelity/quality to get the best results.
However, I just added another knob to the Diarizer class called mer_cos, which controls the speaker merging threshold. The default is 0.875, so perhaps try lowering to 0.8. That should help.
I'll also get around to adding a oracle/min/max speakers feature at some point, for cases where you know the exact number of speakers ahead of time, or wanna set upper/lower bounds. Gotten busy with another project, so haven't done it yet. PR's welcome though! haha
Thanks, `mer_cos` definitely gets me closer. I appreciate that. Yeah, I was thinking providing a param for the expected number of speakers would be nice. I'll check out the codebase and see if that's something I can contribute :).
Yeah would love contributions! Here's a brief overview of how I think it can be done:
Senko has two clustering types, (1) spectral for audio < 20 mins in length, and (2) UMAP+HDBSCAN for >= 20 mins. In the clustering code, spectral actually already supports orcale/min/max speakers, but UMAP+HDBSCAN doesn't. However, someone forked Senko and added min/max speakers to that here (for oracle, I guess min = max): https://github.com/DedZago/senko/commit/c33812ae185a5cd420f2...
So I think all that's required is basically just testing this thoroughly to make sure it doesn't introduce any regressions in clustering quality. And then just wiring the oracle/min/max parameters to the Diarizer class, or diarize() func.
You doing code completion and agentic stuff successfully with local models? Got any tips? I've been out of the game for [checks watch] a few months and am behind on the latest. Is Cline the move?
I haven't bothered doing code completion locally yet, though its something I want to try with the QWEN model. I'm mostly using it to generate/fix code CLI style.
I had some pretty decent but very non-state-of-the-art success with it even cobbled together with LM Studio and VSCode plugins. I'm excited to keep trying it over the next months and years.
As a practitioner I also inherently believe in well written software but as a lifelong learner, things change, and evolve. There is absolutely no reason why software today has to be written like software of yesterday.
There is no need to be so prescriptive about how software is made. In the end the best will win on the merits. The bad software will die under its own weight with no think pieces necessary.
On the other hand, code might be becoming more like clay than like LEGO bricks. The sculptor is not minding each granule.
We don't know yet if there's long term merit in this new way of crafting software and telling people not to try it both won't work, and honestly looks like old people yelling at clouds.
> There is absolutely no reason why software today has to be written like software of yesterday.
I get what you're saying, but the irony is that AI tools have sort of frozen the state of the art of software development in time. There is now less incentive to innovate on language design, code style, patterns, etc., when it goes outside the range of what an LLM has been trained on and will produce.
Personally I am experimenting with a lot more data-driven, declarative, correct-by-construction work by default now.
AI handles the polyglot grunt work, which frees you to experiment above the language layer.
I have a dimensional analysis typing metacompiler that enforces physical unit coherence (length + time = compile error) across 25 languages. 23,000 lines of declarative test specs compile down to language-specific validation suites. The LLM shits out templates; it never touches the architecture.
We are still at very very early days.
Specs for my hobby physical types metacompiler tests:
LLMs aren't like a screwdriver at all, the analogy doesn't work. I think I was clear. LLMs aren't useful outside the domain of what they were trained on. They are copycats. To really innovate on software design means going outside what has been done before, which an LLM won't help you do.
No, you weren't clear, nor are you correct: you shared FUD about something it seems you have not tried, because testing your claims with a recent agentic system would dispel them.
I've had great success teaching Claude Code use DSLs I've created in my research. Trivially, it has never seen exactly these DSLs before -- yet it has correctly created complex programs using those DSLs, and indeed -- they work!
Have you had frontier agents work on programs in "esoteric" (unpopular) languages (pick: Zig, Haskell, Lisp, Elixir, etc)?
I don't see clarity, and I'm not sure if you've tried any of your claims for real.
The last six decades of commercial programming don't exactly bear this out...
The real lesson is that writing software is such a useful, high-leverage activity that even absolutely awful software can be immensely valuable. But that doesn't tell us that better software is useless, it just tells us it is not absolutely necessary.
Software engineers are desperate to have their work be like machining aircraft parts.
It’s a tool. No one cares about code quality because the person using your code isn’t affected by it. There are better and worse tools. No one cares whether a car is made with SnapOn tools or milled on HAAS machines. Only that it functions.
We know there is no long term merit to this idea just looking back at the last 40 years of coding.
Epstein was running what he said was legal tax avoidance system for his clients.
I'm not saying they were breaking any laws in it, but Epstein is a dodgy guy (also he used to be involved in IIRC a Ponzi scheme and was previously sacked from a big firm - red flags galore with this guy) and the scheme worked off asset prices and trusts.
If a bunch of billionaires could manipulate asset prices (selling illiquid assets like mansions and artwork between each other and their trusts) I suspect they could really bring down their tax bills. This would be illegal (I think) but you'd need to untangle a large web of transactions to prove it.
Yes. IMHO, the parts that they really don't want to come out are the financial ties. It's the connections of money (and power and influence) that is being covered up, more then the child sex crimes that are now known.
The tax fraud can get a conviction, since there's a paper trail, and juries are more likely to think "yeah this billionaire might not be a monster but they probably cheated in their taxes".
Epstein didn't rape any children, he had a harem of young women. Some of the women were under 18, but all of them were old enough to be held responsible for their part in the crime, that is, being prostitutes. Many even have admitted to comitting "child" sex trafficking by inviting girls from their high schools to be prostitutes.
Of course the perpetrators of child abuse would be adjacent to child abuse material. That's why the material needs to be indexed, to make sense of it and round up the rest of the perpetrators.
Herding goats doesn't solve the interesting technical problem I'm trying to solve.
Point is: if that problem is solvable without me, that's the win condition for everyone. Then I go herd goats (and have this nifty tool that helps me spec out an optimal goat fence while I'm at it).
> Point is: if that problem is solvable without me, that's the win condition for everyone.
The problem is solvable without you. I don't even need to know what the problem actually is, because the odds of you being one of the handful of the people in the world who are so critical that the world notices their passing is so low, I have a better chance of winning a lottery jackpot than of you being some critical piece of some solution.
I completely disagree - I think it’s the other way around.
Solving the problem - no matter what problem it is - is extremely dependent on you and every single human being (or animal for that matter) is a critical piece of their environment and circumstances.
reply