This project has moved. For the latest updates, please go here.

Equalizing wav files... will it increase accuracy?

Sep 7, 2014 at 6:12 PM
Is there any need to equalize the wav files or does the program do its own filtering?
I was thinking of of boosting just the vocal range (300-3000 Hz), but if the program already does this, it may just make it worse.

Great program!
I've always dreamed of using voice recognition to create subs, but you're the first to actually do it, at least to my knowledge.
Congratulations!

Andor1999
Coordinator
Sep 7, 2014 at 6:23 PM
Edited Sep 7, 2014 at 6:24 PM
Hi, SubtitleTools won't change the input .wav files. This is the best setting for Windows speech recognition engine, using AoA Audio Extractor:

Image
Sep 7, 2014 at 6:50 PM
Hi, VahidN.

I'll definitely give AoA a try.
I used LameXP and have only tried one recognition so far.
It did a terrible job of guessing the words, but it was an old British documentary with terrible sound, so I'm not surprised.
It did do a pretty good job of timing. That's something right there.
Now that I know your program can benefit from pre-processing the audio, I've got some experimenting to do.
Audacity might be just the app. I'll post results if I find something useful.
By the way, I'm running Windows 8.1 64 bit on an i7 and SubTools 2.5 has worked fine so far.

Thanks for the quick reply.

Andor1999
Coordinator
Sep 7, 2014 at 10:01 PM
For the En-UK, it's very important to select the related speech recognition engine.
There was no En-UK speech recognizer installed on my system. Here is its direct link to download for the Win8.1 - X64:
lp_55c5ff4da1d83ae6f4ebd1f8c020cb80ceda7da8.cab
After download, rename the .cab file to .mlc and double click it. Run through the wizard to install the language.
Method 2: Install Language Packs for Multilingual Support
Now you should see the En-UK here:
Image
Sep 7, 2014 at 11:15 PM
Thanks for the link.
I installed the UK cab and now I see both the US and the UK version in the Speech Engines drop down.
It didn't make much difference though. Accuracy is still very poor.
I experimented for several hours with equalization, normalization, amplification and other tools in Audacity.
Nothing really improved accuracy. In fact, most made it worse.

Do you have a wav file you could share that has high recognition so I can test my setup?
If you could also share the srt so I'll be able to compare it with my results.

I don't know if the forum is set up to share files like this.
If not, I can send you my email address and you can send them with something like FileMail.

Thanks for your help!

Andor1999