Help with LF_Rolloff_for_speech

[kozikowski]

If the work is under 25MB, it can be sent over plain email.

Koz

[Jan B.]

Well, I am at a standstill. I don't have any unprocessed (or unpatched or raw) audio to upload. I processed it all, five chapters, after uploading an audio sample, getting feedback on the quality, and applying the suggested fixes. I used Noise Reduction, Equalization, and Normalization, with numbers kindly supplied here on the forum presumably based on the sample, and then the LF_Rolloff, and yet the ACX analysis showed the work fails the test. I need to know if the work already done can be fixed, before going forward, and shouldn't that be able to be known by the numbers in the ACX analysis rather than by submitting new raw audio?

[kozikowski]

shouldn't that be able to be known by the numbers in the ACX analysis

No. It only tells us the train is late. It doesn't tell us why or the proper method of fixing it.

Software doesn't understand content and it doesn't know what a voice is. In the case of loudness, all it does is sweep all the blue waves into a bucket and if the bucket doesn't fill up, it fails.

One of the ACX failures is "Overprocessing." That's where someone smashes their voice into ACX Conformance even though the process makes them sound funny. You are also producing good quality theater/radio which is a very close cousin to audiobook production.

Anyway, yes, post one of the adjusted chapters and tell us where you put it and what failed.


It's a terrifically good idea to save archival copies of your recordings before processing. I still have original shoots from two years ago and I have gotten calls from producers wanting them.

Koz

[Jan B.]

Yes, I see the point of archived unprocessed copies now!

I'll post seconds of work, not the whole chapter, as you originally asked. I appreciate it. It might take me a couple days, family stuff.

[kozikowski]

The reason I asked for a chapter is to see how our tools work over the long haul. Anybody can design a tool, for example, that takes a second or two to do its job over a short performance, but takes an hour on a full show. That's a nasty surprise if you're on deadline.

Also, we had one performer who submitted a test recording and no matter what we did, our sound readings didn't match his. Turned out, he was measuring the full show and we were only measuring a short segment. The voices didn't exactly match and that's not a shock to anybody — unless you've never done this before.

This kind of thing could very easily be the difference between the tool set and corrections we told you and the ones you actually need.

You can have theatrical oddities, too. Drop a pencil or do anything to create a sharp, staccato noise in the show and that will throw all the measurements off. This oddity can be built into the story which can really make you crazy. A recent poster wanted to know how to speak a gun blast. It appears in the script. It says so, very clearly, right there. But you're not allowed to get loud all of a sudden in the middle of natural dialog. That could very seriously throw off the readings and corrections.

You'll have all these problems and variations nailed by the third or fourth book.

Koz