[HELP] Converting audio to text

[PsychoΔelica 648]

Hi guys, I was wondering if anybody knows a software that can convert audio speech to text? I've been struggling with editing something I recorded on my mobile and I got really frustrated. Though I don't expect everything on that recording to be decipherable, but that would make my life a lot easier if I could find an app or program that could do that for me. So any help is appreciated.

[beni 2149]

Would this site be of any help?

It seems like most, like the above, you have to pay for... I had found a Chrome App (simply called Transcribe) but that also seems to ask for a fee. The best I could find is this one. Looks like that's free but I could be wrong.

I can continue searching later in the day but these seems to be the top ones when searching.

[PsychoΔelica 648]

Thanks hon, but none of these links is useful. The free transcribe site is not a solution because I still have to do everything manually (I was actually looking at it earlier), and that's not what I was looking for. Looks like there are no shotrcuts for me if I don't want to pay for such a service. Ah well, manually it is then.

 

Or maybe I still haven't tried everything

[Zeus 7997]

For now, no.

Transforming speech to text is a very difficult challenge for computers to handle. Think of it at a more conceptual level to see the whole picture: a computer, which has no sense of hearing, has to interpret your sound, apply any relevant mathematical corrections, and then pick the correct word out of an arbitrary language. Then there's problems like background noise and regional/local accents which can corrupt the message. Finding an algorithmic or mathematical computation to describe what an accent is, how to identify one, and how to write a program to detect them are things people write doctoral theses for. Then there are challenges to overcome such as storing this information in a database (it would be absolutely huge, probably in terabytes or even petabytes if you want to have it work with most of the English language), machine learning, real-time processing, web services, and complex mathematical analysis (I can think of the Fast Fourier Transform off the top of my head as an absolute must in the signal-processing part of this imaginary algorithm).

All is not lost however. Programs like this do exist, but it's not going to do what you want it to do. You have to talk to these programs in order for them to interpret your speech as text, and they're not very good. I have a type of program like this on my smartphone and I barely use it because it gets so many words wrong I have to go back and edit the message anyway. If you're introducing a layer in the forum of a recorded message that you have trouble distinguishing, don't expect the program to get most of it right. I would actually be surprised if the program on my phone translated anything for you because it has a habit of giving up when I speak clearly to it, but even if it did it would butcher the whole thing.

I'm sure the government has already developed some prototype that can just analyze audio and return text for you and it's how I assumed they could retrieve conversations from garbled audio clips (think low-quality phone conversations between terrorists that they tap into). If such a program has made it out of military hands into the consumer market, it would be very expensive and not something you could acquire without struggle. I'll admit I didn't look at all for a program like this to exist, but I'm relying on my expertise and intuition which heavily suggests it doesn't. If someone manages to come into this topic with a link to a freeware (or even free trial) program that transcribes audio to text with a high success rate, I'll eat my hat. Two hats if it can do it in multiple languages.

This is one of those lovely cases in computer science where describing the problem is easy and the solution is insanely hard.

 

[PsychoΔelica 648]

Thanks for giving me hope Zeus Well I didn't really expect to find a magic solution but I didn't think it was this bad. Suppose I'll wait for a spy software to be cracked and put on a dl site Btw I'm working with Hebrew, not with English.

 

As to transcribes, haven't found that very effective so far.

[beni 2149]

I'm sorry this isn't working out. ;_; But thank you Zeus for the explanation!