I personally tested all the STT models for my real-time translator (https://fliptalk.ai). From language detection and accuracy in a noisy environment to the most important point: latency.
At the moment, Soniox STT v5 is definitely the best, and I'm impressed by its performance. It's good that Google released Gemini-3.5-Transcribe, and it beats every other model on accuracy, but it definitely needs a bit more work on latency, which is the most important factor for STT apps.
For me, Gemini-3.5-Transcribe actually has slightly lower latency. Kudos to Soniox for both paying their competitor and letting them win. But yes, Soniox is much cheaper.
Interesting, I uploaded a voice recording from a meeting I had recorded with a relatively cheap microphone.
Soniox came out really good. OpenAI started getting some things very wrong and even introduced some German. Google did okay but cut off the start by several seconds.
What's Soniox doing (left most) that's making it so good ? It was also the only one that could distinguish between the speakers.
Oh, thanks for pointing me to Soniox. It is really good. Would also pick up the words with different languages, identify and output in the right language.
Looks interesting. It was much faster too, but that I cannot say much since it was on their own website.
> latency, which is the most important factor for STT apps.
Perhaps latency is more important than accuracy for a real time translation app (I actually disagree with this - imagine e.g. the hilarity when requesting "a new display" being translated as "a nudist play"), but certainly not for all applications. My pet app transcribes personal voice notes to self, it could run all night.
I agree. For my use case, I chose to prioritize latency over accuracy, but it's always difficult to find the right balance between the two. There is no easy answer.
Realtime + Voice AI usecases is where latency is most important. I use Handy on my desktop and i can tolerate a latency of a few seconds every now and then. Your P99 should on TTFB should be really low to compete for voice ai realtime
Simply put, without VC funding, these services would cost hundreds of dollars per month, and nobody would pay those prices. Companies like Anthropic can offer $20 plans precisely because they have venture capital backing.
Consumers will always gravitate toward the more affordable option.
I think most people do not realize that the real secret is just doing the thing most of the time and not overthink the times the thing could not be done.
And many don't understand that time is limited but energy and focus also is.
It does not matter that stuff are planned/schedule or that there are blocks of "free" time, you really can't maximize efficiency for everything, and it would not make a significant difference anyway.
The key is strong focus on what matters and letting everything else in the background.
This is why very often I get absolutely stunned at how much inefficient crap people put up with in their life, of course at the end of the day they have no real time/focus power to do anything worthwhile.
The task of managing your habits can also become another thing to put energy into. When it's frictionless it helps get stuff done, but when it's complicated you end up organising and moving information around instead of becoming more effective.
If SEO's a concern, then there's nothing you can do about the LLM bit. Either put the website behind a login, or deal with robots sucking everything up.
Pre-LLMs, there were already many websites showing scraped versions of other people's sites, and ranking higher in search results.
> Ruby's poor performance hobbles it even in places it's supposed to be good (twitter was rewritten to Scala over rails).
All performances are relative. Ruby or Python is very performant for 99% of websites, generally the bottleneck is the database.
Rails has enabled Twitter to develop and iterate faster. But now that the social network has to manage several tens of millions of users, the 200 req/s of Rails are no longer sufficient.
And Ruby in 2013 didn't have the same performance as in 2023. Hardware no longer had the same performance, too.
I still remember the day I met John Warnock as if it was yesterday.
John's humility, despite his monumental success, taught me to always keep my feet on the ground. His vision inspired me to push boundaries in software development. But above all, it was his passion that kindled a fervor within me to harness technology for change.
John Warnock's passing is a massive loss for the tech world. My heart goes out to his loved ones during this time. His legacy, however, will continue to shape the industry, inspiring countless others just as he had inspired me. Goodbye, Mr. Warnock. You will be missed.
I've created Claper [0], an open-source tool for real-time interaction with a public. You have multiple ways to monetize an OSS, starting from donation, sponsorship to consulting or SaaS.
Personally, I chose the last one with a cloud version. I like this model, people paying for the SaaS support the project and have, in return, a hosted version of the OSS, and so contribute to improve it.
Are you sure about the fixed fee? Their website says it isn't.
> We do not have monthly fees, set-up fees, integration fees or closure fees. We do have a minimum invoice depending on industry or business model. Please speak to a member of our sales team for more details.
At the moment, Soniox STT v5 is definitely the best, and I'm impressed by its performance. It's good that Google released Gemini-3.5-Transcribe, and it beats every other model on accuracy, but it definitely needs a bit more work on latency, which is the most important factor for STT apps.