Загрузка видео...
Не удалось загрузить видео
Introducing VIVA 2.5, with our most advanced models yet. Krisp's Voice Isolation has been running in production for 2 years, across 1B+ mins of voice AI conversations a month, improving WER by focusing on primary speaker only. As more teams ran voice isolation in front of their STT, we... show more
25,958,617 просмотров • 1 месяц назад •via X (Twitter)
Комментарии: 34
@krispHQ Looks like shit pal
@krispHQ Genuinely, what's the benefit of this technology?
@krispHQ I fucking hate ai
@krispHQ Still looks fake, the uncanny valley vibe will never leave AI slop
@krispHQ I dont. I'm old school i can still think for myself. Thats what a brain is for 😉
@krispHQ who needs 2 years of data to figure that out lol
@krispHQ its so HD that it went from looking convincing back to looking uncanny lmao
@krispHQ give me back my water
@krispHQ I didn’t know Vauxhall fitted a 2.5 in the viva 🤷♂️😀
@krispHQ Let’s get rid of this Ai slop and make em lose the money the spent to corrupt our minds and importantly our kids minds personally I don’t want my kid to use Ai to learn as it legit makes em dumb typing everything in and nog learning anymore
@krispHQ Looks and sounds like AI slop
@krispHQ super intéressant de voir l'évolution de l'isolation vocale en prod. vous mesurez aussi la latence ajoutée par le traitement ?
@krispHQ why don't u work on something that actually helps humanity instead of perfecting make-believe "customer service agents" that end up referring to actual human beings to answer the simplest questions —but then they don't even know their own business's customer service hours #Fail
@krispHQ We don’t want any of this
@krispHQ This should make voice agents feel much more reliable.
@krispHQ This shit is the reason glaciers are melting and leaves are falling in August. Let actors work ffs
@krispHQ Yeah, I had a Vauxhall Viva in the 70s, it wasn't that big though 😕
@krispHQ Hey A.I engineer here and I worked on Viva and I can confirm it was trained on massive amounts of porn especially gay porn. I said "Davit why do you need all this for model training?" And he said it was very necessary to have terabytes of porn sent to him personally.
@krispHQ I love him. ( But not as much as Liam Gallagher !! )
@krispHQ Why the fuck are people striving to achieve human like video to a point they become indistinguishable from reality. Are those fucks blind, egotistical or too stupid to see the danger? Or the danger is the goal?
@krispHQ The 3.5x smaller model is just as exciting as the accuracy gains. Better efficiency opens the door for many more deployments.
@krispHQ Love how you fixed the over-aggressive isolation issue without hurting clean audio. Clean progress.
@krispHQ Bora
@chesny @krispHQ 🔼 Less guessing. More analysis.
@krispHQ But the CFF is way too high. Fix it.
@krispHQ The US remake of Peep Show looks terrible.
@krispHQ Better voice isolation means cleaner AI conversations
@krispHQ 2.5 É um vinho da beira interior, ( Caria )ótimo para acompanhar, caça, queijos e carnes vermelhas.
@krispHQ I read a lot of launch posts. Almost none admit the previous version had a regression. @krispHQ just did, and shipped the fix with receipts: 46.4% fewer word errors across 10 STT engines, no harm on clean audio. That is how you earn trust in voice infrastructure.
@krispHQ 1B minutes of convo, impressive! Any plans for multi‑speaker support?
@krispHQ @BrainPathio
@krispHQ Woooow 🚀
@krispHQ This sounds like a pretty big upgrade! I like that you're focusing on improving accuracy, especially in noisy environments.
@krispHQ what kind of impact did it have on noise levels in meetings?
![Google presents AudioPaLM: A Large Language Model That Can Speak and Listen paper page: introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation. AudioPaLM inherits the capability to preserve paralinguistic information such as speaker identity and intonation from AudioLM and the linguistic knowledge present only in text large language models such as PaLM-2. We demonstrate that initializing AudioPaLM with the weights of a text-only large language model improves speech processing, successfully leveraging the larger quantity of text training data used in pretraining to assist with the speech tasks. The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training. AudioPaLM also demonstrates features of audio language models, such as transferring a voice across languages based on a short spoken prompt.](https://image.24vids.com/tw-1672048076469698561/ext_tw_video_thumb/1672047839378300928/pu/img/MuoOBsAyUd-HAI7D.jpg)
![Remember when AI couldn't draw a hand? Seven fingers, knuckles pointing backwards. And the AI spaghetti videos. That was three years ago. Images are done now. Video is close enough that you scrolled past AI ads this week and clocked exactly zero of them. Code writes itself and there are like 40 coding agents. AI voice spent that entire stretch sounding like the lady voice in a 2014 GPS. Flat, evenly spaced and every sentence landing with the same weight, like it's reading from a phone book. Here's why it stayed broken. Bad images are funny. You screenshot the seven fingers, it goes viral for being bad, someone fixes it. Bad audio is just boring. It doesn't fail spectacularly, so it never got that pressure. The bigger problem was the scoring. The whole industry graded AI voices on whether you could make out the words. So the models learned to over pronounce everything, hitting every syllable like a newsreader. Perfectly clear but robotic. Everyone was chasing a score that had nothing to do with sounding human. Meanwhile a small open-source team was doing something harder. Their lead researcher, an ex-NVIDIA engineer, went all in on an approach the rest of the field had written off. Two years early. No funding announcements or launch tour. He just put the whole thing on GitHub for free. It's sitting at 50,000+ stars now. Then they ran the test everyone else avoided. For 10 days they piped real users through their model and every big competitor with the listener never told which was which. Thousands of real people, real scripts. Whichever voice you actually preferred, they logged it. Theirs came out on top. It beat ElevenLabs about 6 times out of 10, head to head. It beat OpenAI's voice model 8 times out of 10. The gap was widest on the breathing, the pauses, the little hesitations, which is exactly the stuff that makes a voice sound like a person instead of a machine reading. They ran on real users rather than a lab, which is more than most of these claims can say. That's Fish Audio. This week they shipped S2.1 Pro: - Clone anyone's voice from 15 seconds of audio - Fast enough to hold a live conversation - 83 languages, one model - Type [whisper] or [sigh] mid-sentence and it does it - Around 70% cheaper than ElevenLabs - Free to download and run yourself Voice was the last thing on the list. Around 20 people with a free repo got there before other billion dollar companies did.](https://image.24vids.com/tw-2082409789385224219/amplify_video_thumb/2082408743611113472/img/Iub3c6E4muQa_yeC.jpg)
