@lucasmeijer You also get better quality from forced alignment since it knows what’s being said as you already have the script. I’ll show you an example later today
English
Fung
474 posts
Fung
@Funggx
never hurts to learn more | undergrad ML Research & SWE
Katılım Ağustos 2024
164 Takip Edilen41 Takipçiler
@Funggx I have low volume. I only care about quality not money.
English
I really want these guys to be great, but I'm 30 clicks after seeing this tweet, and I still have not found the actual thing they shipped! I have production apps that could use this, and I give up because it's always such a shitshow to use anything google ai.
Logan Kilpatrick@OfficialLoganK
Introducing Gemini 3.1 Flash TTS 🗣️, our latest text to speech model with scene direction, speaker level specificity, audio tags, more natural + expressive voices, and support for 70 different languages. Available via our new audio playground in AI Studio and in the Gemini API!
English
@lucasmeijer Performing STT just to get timestamps is usually very expensive compute/time wise. I haven’t worked with Dutch before, but you should use forced alignment given that you have the text already, it’s way cheaper. I can build and show you a sample of it
English
@Funggx I just tried Inworld TTS. Its dutch quality is horrrible, getting way better results with that new gemini thing.
I guess I just need to run the generated audio through a transcriber again to get timestamps to drive subtitles
English
@lucasmeijer Gemini TTS is not very good for this. Inworld TTS API (better quality imo) gives you word/character timestamps. But if you need to use Gemini TTS, you can perform forced alignment (github.com/lukerbs/forcea…) to create the SRT file. If you want more help, feel free to send a DM :)
English
@Funggx what do you recommend if I want to drive subtitles with the audio track and I need word level timecodes?
English
this is literally unusable holy shit. he's running a thousand git commands am i just supposed to accept a thousand times
taoki@justalexoki
why can't i bypass permissions in the claude app. i'm just sat here pressing accept a thousand times. this is retarded
English
That is the real tier list ive seen so far.
Ara Song@choehau_ara
Real dudes uses firefox and harden it.
English
@PabloPeniche @tenobrus @sheherenow_ No monthly plans available? Would love to buy but only see the annual options
English
@tenobrus @sheherenow_ happy to give you both discounts, DM me!
English
holy shit just replaced whispr flow with this and it's instantly a 10x better experience
Aqua Voice@aquavoice
Aqua Voice is now live for iOS. It's a premium voice keyboard for every app on your phone.
English
@msfeldstein @kr0der cursor cloud agents are peak, I love how they record the testing. Sadly they cost a ton to run and don't work with cheaper models
English
@kr0der Ive been loving cloud agents recently on web. I've been doing CLI development and you can use the terminal right in the browser so i just dont have to do anything locally, parallelization ++
Like this is all i need to build with an agent and try out the results
English
what do you guys think about cloud agents?
i don't really understand the hype in terms of using them for coding - they become a LOT slower, more expensive, and the UX to get them working perfectly isn't the best compared to local agents
i still prefer local agents for coding by a lot and pretty much only use cloud agents for a quick fix in Slack
English
does this mean GPT 5.3 Codex is actually better at coding than GPT 5.4? 👀
Tibo@thsottiaux
@Its_Nova1012 GPT-5.4 - Best results, highest cost. Well rounded. GPT-5.4-Mini - Cheapest option. Fast and accurate, will notice occasional things that are too hard for it and you can switch to GPT-5.4. GPT-5.3-Codex - Code freak. Can use if you only care about generating code.
English
@prathamdby Have you ever ran into any awkward model behaviors? This seems like gold, especially with something like a copilot sub on top of it
English
I just made a short video showing how to do this with the CLIProxyAPI tool. It's a very useful and easy-to-use tool, but it can be hard for people who have never used it before. In this video, I showed you how to use any OAuth subscription in Cursor AI.
Links below ⬇️
Fekri@fekdaoui
hey @cursor_ai i’d love to be able to use my openai codex sub in cursor’s native agent window you could even make it available only to active cursor subscribers so people don’t just free-ride cursor ux + codex sub usage limits would be elite
English
English
will be streaming high perf cuda i guess since it's a draw. reaches more people and probably more helpful :)
you wouldn't believe how much inefficiency exists in open source codebases/tooling unless you very actively read training and inference frameworks. all the main ideas are basically out there but very scattered
we will push the boundary and optimize the fuck outta consumer and eventually datacenter gpus.
no clankers, no optimization frameworks, no dsls, no search-based compilers, just raw perf and lowering memory requirements from first principles
Aryan V S@aryanvs_
A or B? bird app does not let you put poll in quote tweet :/
English
alright i'm officially gonna try friends of friends only replies for at least 2 weeks and see how it affects my experience. will reconsider then.
but before the gates slam shut, it's Lowbie Appreciation Thursday! introduce urself in the replies and tell me why i should follow u!
Tenobrus@tenobrus
strongly considering making all my posts "friends of friends replies only" for a while. i follow 2.5k people and they collectively follow an assload more, so its really not too strong a filter, but it'd keep out bots and random groypers. downside is it blocks brand new lowbies...
English
Love that they are comparing with 2 generation old Luma models. @Designarena lets try with Ray3.14
Design Arena@Designarena
BREAKING: Grok Imagine by @xai takes 1st overall on Multi Image to Video Arena, with an overall Elo of 1342. The team's debut reference image to video model establishes a new Pareto frontier for Preference vs. Speed with an average generation time of 58.9 seconds. Huge congrats to the @xai team for this achievement!
English
i can recognize this place from anywhere lol
WeenUpdates@WeenUpdates
JASONTHEWEEN GOT FLEXED ON AFTER SAYING HE DIDN’T HAVE A DOLLAR 😭😭
English
@iwanttobeinmars thats like thinking about ur grandma when having sex
English