ASR = automatic speech recognition. I think this is the same as STT, speech to text?
It seems to want an NVidia GPU. I didn't look closely to find out if it can run usably without one. For me, usable might mean 10% of realtime, i.e. an overnight run on my old x86 box for 1 hour of transcription. Of course faster is always nice. YMMV.
Almost all ML models come with reference python code that will run on most architectures.
If you want an implementation optimized for a certain architecture the cost is between 10-30USD of opus tokens and is oneshottable: ("implement Nvidia nemotron 3.5 transcribe for the intel npu on this machine. Use the hf transformers implementation as reference. The finished artifact takes pcm on stdin and utf8 chunks on stdout".)
You can substitute your old intel CPU by modifying the prompt by replacing "intel npu" with "the target cou on this machine ... Your task is complete when the model decode performance is 80% of the memory read speed of this machine"
Thanks, that's interesting, I hoped maybe it would use pytorch or something. But what I'm wondering is whether the result would run at usable speed. I don't currently use anything like opus and only want to use open weight models, and at the moment I don't have a GPU machine, though I guess I can spin up an hourly VM with one sometime.
> ASR = automatic speech recognition. I think this is the same as STT, speech to text?
I'm surprised no one corrected you. There's no speech left when it's converted to text, while, as you stated, the other one is Automatic speech recognition :-)
I don't see the need for an even more powerful Pi Zero. I'd say just have a modernized version of the original $5 single core Pi Zero, or ok the wireless version was good too. Maybe even have a 128MB config (original one was 512MB before the DRAM shortage) to keep the price down. Model A same same, who needs it if it's not a lot cheaper than the B.
I do like the Pico series and it would be cool to have one of those with memory protection.
It's because of IP reputation and spam blocking. You can't start with a "clean" address and send mail. You need to spend some time on "address warming" to build reputation for the address before the big providers will accept mail from it.
It's basically impossible to not use gmail. Even if you don't have a gmail account or address yourself, the person you're emailing probably has it, so your messages gets sent to a gmail server and gets scanned by them. What I want to know is whether Google uses the contents to profile YOU.
Is it impractical to never send a message through Google? Probably, depending on the circles you communicate in. But that's not the same thing as saying it's impractical to avoid Google for most of your messages, and certainly not the same as saying you can't avoid GMail for anything.
My main email is not GMail, but I forward it to a GMail account to get free spam filtering, free backup, and free search. I encourage everyone to do the same.
You don't get spam if you use aliases everywhere. The few that actually get spam can be turned off.
Like I recently had to turn off alias for made4men.no. srsly, signing me up for spam and not removing me from the spam list even after i unsubscribed. You can tell it's a swedish shop that just happens to own a Norwegian domain.
I actually had to do the opposite. I have a filter that catches all messages marked as spam and unmarks them as spam. VERY rarely there's a spam message caught in there. Much more often sth super important that just would have landed in spam if not for my genius spam filter.
This is a self-fulfilling prophecy. If everyone reads "it's impossible to avoid gmail, just use gmail", then yeah, gmail prevails.
But it's plain false. YOU can decide against using gmail. I don't use gmail. I sometimes talk with people who use gmail, but more often than not, I don't. Stop being fatalistic.
I wonder whether they were trained for this on purpose, to accelerate the general public turning on each other and handing the world to the broliarchs. Added: alternatively, I wonder if the training materials have a sociopathic bias because of the kinds of places they came from.
Mass isn't close to being conserved. If an electron and positron annihilate, they become two massless photons, for example. You don't need to invoke an expanding universe.
In an expanding universe, energy isn't conserved. (Unless we do the trick we always do with energy and just define it to be conserved https://youtu.be/lCgKTpIzjRk?t=174 )
I don't know exactly why you think this applies to charge conservation. The U(1) symmetry is where charge conservation comes from and it isn't affected by an expanding universe.
There is "rest mass" and "relativistic mass". The GP is talking about "rest mass" that is not conserved even in simple examples.
Most modern books try to eliminate "relativistic mass" because it's misleading, for example is good to calculate acceleration perpendicular to the movements (for example in a cyclotron) but it's not good to calculate acceleration in the direction of the movement.
So most modern books they call the same idea just "energy" and use that "energy" to calculate gravitation. "Energy" is conserved in special relativity and almost conserved in general relativity.
But the "relativistic mass" is useful in some problems and is widespread in popular culture and don't ask about the weird parts of the mass of the proton, so it's difficult to kill the idea.
reply