Most probably all VCs will go out of business as the cost of creating software companies approaches zero. The need to create an army of software developers is no longer needed.
In such a hypothetical world, it would actually be much easier for us to fund companies simply because now the only thing we are funding is just sales, demand gen, and projected compute.
We provide funding so businesses that are the right fit can scale out the functions that they need. In some cases it's expanding engineering, in other cases it's expanding sales and demand gen, and in other cases is to subsidize a major purchase such as cloud credits or GPUs.
I don’t think it is about resources. Growing up in 3rd world country I have zero resources. It is the drive of student and higher standards from parents and teachers that matters. Everybody is just getting soft.
Just made a TTS tool based on Kitten TTS, fully browser based, no Python server backend: https://quickeditvideo.com/tts/
A tts model of this size should be industry standard!
The people calling it "OK" probably tried it for themselves. Whatever model is being demoed in that video is not the same as the 25MB model they released.
It doesn't sound so good. Excellent technical achievement and it may just improve more and more! But for now I can't use it for consumer facing applications.
Speech speed is always a tunable parameter and not something intrinsic to the model.
The comparison to make is expressiveness and correct intonation for long sentences vs something like espeak. It actually sounds amazing for the size. The closest thing is probably KokoroTTS at 82M params and ~300MB.
The voices sound artificial and a bit grating. The male voices especially are lacking, especially in depth: only the ultimate voice has any depth at all, while the others sound like teenagers who haven't finished puberty. None of the voices sound quite human, but they're all very annoying, and part of that is that they sound like they're acting.
The only real questions are which Chinese gacha game they ripped data from and whether they used Claude Code or Gemini CLI for Python code. I bet one can get a formant match from output this much overfit to whatever data. This isn't going to stay up for long.
Impressive technical achievement, but in terms of whether I'd use it: oof, that male voice is like one of these fake-excited newsreaders. Like they're always at the edge of their breath. The female one is better but still someone reading out an advertisement for a product they were told they must act extra excited for. I assume this is what the majority of training data was like and not an intentional setting for the demo. Unsure whether I could get used to that
I use TTS on my phone regularly and recently also tried this new project on F-Droid called SherpaTTS, which grabs some models from Huggingface. They're super heavy (the phone suspends other apps to disk while this runs) and sound good, but in the first news article there were already one or two mispronunciations because it's guessing how to say uncommon or new words and it's not based on logical rules anymore to turn text into speech
Google and Samsung have each a TTS engine pre-installed on my device and those sound and work fine. A tad monotonous but it seems to always pronounce things the same way so you can always work out what the text said
Espeak (or -ng) is the absolute worst, but after 30 seconds of listening closely you get used to it and can understand everything fine. I don't know if it's the best open source option (probably there are others that I should be trying) but it's at least the most reliable where you'll always get what is happening and you can install it on any device without licensing issues
anyone else wants to try sherpaOnnx you can try this.. https://github.com/willwade/tts-wrapper we recently added in the kokoro models which should sound a lot better. There are a LOT of models to choose from. I have a feeling the Droid app isnt handling cold starts very well.
Somebody should create a AI interviewer for VC funding. VCs are swamped with so many funding requests. All the founders should first convince AI why they need funding.
And it’s not even because you don’t want to. It’s just because that’s how things work. I spent years talking directly to users and then I started working for a multinational, and I haven’t seen a user in 7 years…
