Ralf

And we have the big advantage, if you push it from ultrasound, you also have all the tracks nicely separated and you can put every single one into recognition. That means, when people talk to each other, it doesn't matter to the WSFAB model at all, because it analyzes every track for itself and then throws the whole thing back into the timeline.

And we have the big advantage, if you push it from ultrasound, you also have all the tracks nicely separated and you can put every single one into recognition. That means, when people talk to each other, it doesn't matter to the WSFAB model at all, because it analyzes every track for itself and then throws the whole thing back into the timeline.

That means, the recognition quality will be much better than with services where I just have a stereo sum. So I'm analyzing the voices of the whole podcast.

That means, the recognition quality will be much better than with services where I just have a stereo sum. So I'm analyzing the voices of the whole podcast.

So that's going to happen. We all have the components together. Because the result, when we have the full texts, then we can write them in the timeline. I already have a demonstrator ready for that. Then we also use a special tagging function from Reaper, that you can attach a time stamp within an audio track. And then you really have in the editing program

So that's going to happen. We all have the components together. Because the result, when we have the full texts, then we can write them in the timeline. I already have a demonstrator ready for that. Then we also use a special tagging function from Reaper, that you can attach a time stamp within an audio track. And then you really have in the editing program

over the wave form then the recognized full text and can search and filter and with that we can of course also in the audiogram very simple render so text rendering is there trivial and the timing information when which word is spoken we have them too because it's really at the right place in the timeline so that will come

over the wave form then the recognized full text and can search and filter and with that we can of course also in the audiogram very simple render so text rendering is there trivial and the timing information when which word is spoken we have them too because it's really at the right place in the timeline so that will come

And maybe interesting and also a little bit of a transition to your AI topics that will come soon, Tim. I have just before the subscribe, how long has it been now? Four weeks, five weeks? Not so much, a little longer, but about a turn. I did tests before, because I always think, okay, if we already have a complete transcript of the show.

And maybe interesting and also a little bit of a transition to your AI topics that will come soon, Tim. I have just before the subscribe, how long has it been now? Four weeks, five weeks? Not so much, a little longer, but about a turn. I did tests before, because I always think, okay, if we already have a complete transcript of the show.

Then you have to be able to do exciting things with LLMs as well. Keyword, automatic generation of chapter marks, for example. Or write me a summary. Or look for the links again, about which we are talking here. Who took the most effort? That's what I'm getting at now. And before the subscribe, it was a hopeless undertaking. Because this...

Then you have to be able to do exciting things with LLMs as well. Keyword, automatic generation of chapter marks, for example. Or write me a summary. Or look for the links again, about which we are talking here. Who took the most effort? That's what I'm getting at now. And before the subscribe, it was a hopeless undertaking. Because this...

token window of the LLMs was simply too small to handle a 4-hour broadcast of Freakshow. That means I always had a broadcast transcript, which we also offer here for Freakshow. Every broadcast comes from Auphonic over in that part and the recognition is also very good there. And I always throw that in and say, make an input, make a

token window of the LLMs was simply too small to handle a 4-hour broadcast of Freakshow. That means I always had a broadcast transcript, which we also offer here for Freakshow. Every broadcast comes from Auphonic over in that part and the recognition is also very good there. And I always throw that in and say, make an input, make a

And the LLM can do that intellectually, but so far they just couldn't deal with these amounts of text. So it doesn't matter what they promised you, how big the token window was supposed to be, there are always these effects.

And the LLM can do that intellectually, but so far they just couldn't deal with these amounts of text. So it doesn't matter what they promised you, how big the token window was supposed to be, there are always these effects.

So the first three pages, they were processed quite well and then the whole thing went down into such a large area of attention and then at the very end it goes up a bit again, then one or two markers are set correctly again, but the result was completely unusable. And now I tried the same thing again two weeks ago. And look, it works.

So the first three pages, they were processed quite well and then the whole thing went down into such a large area of attention and then at the very end it goes up a bit again, then one or two markers are set correctly again, but the result was completely unusable. And now I tried the same thing again two weeks ago. And look, it works.

That means now with the current Claude Sonnet model, the 3.5, which I have already praised in another context in the broadcast. That will also be praised again today. We will price that further. As well as the O1 Mini can now suddenly... from one week to the next with the amount of text of a complete freakshow. And then Tim and I had a little chat where I happily reported it.

That means now with the current Claude Sonnet model, the 3.5, which I have already praised in another context in the broadcast. That will also be praised again today. We will price that further. As well as the O1 Mini can now suddenly... from one week to the next with the amount of text of a complete freakshow. And then Tim and I had a little chat where I happily reported it.