How to fix this crappy dialog?

musman wrote on 6/29/2004, 1:15 PM

Nearly finished VAAST Sound Forge dvd and have learned a lot. Now I'm trying to figure out how I should use these tools and Vegas for a short film I shot a while ago and could use some suggestions. It was shot and audio recorded into a Sony pd150 camera which does not have great audio. The biggest problem is the audio recording is way too low and bringing it up to a reasonable level introduces a lot of noise and hiss. Of course, there are also moments when the audio is much louder, they're mixed in between all the bad stuff.
I've experimented a lot with Vegas and Sound Forge but was wondering:

1- Should I be aiming for dialog to peak at around -3db with the rest of it as close to this as possible?

2- What is the best way to get my audio uniformly near the -3db? Right now audio peaks anywhere b/t -50 and -30 db, so it's got a long way to go. Normalization to speech first brings it up a lot but there's still a big variance b/t the louder bits and the rest of dialog. Should I use compression first?

3) Anyway I bring things up introduces a lot of noise. Should I use hiss cutting EQ settings before bringing things up or compressing? Should noise reduction be the last thing I should do?

Sorry for the long post, but I'm a newbie with sound and am trying to learn the best ways to use Sound Forge. Any help is appreciated!

Comments

JJKizak wrote on 6/29/2004, 2:05 PM

I don't have any solutions for your level problems but if you had the Sonic Foundry Noise Reduction pluggin the steady noise would be a
piece of cake to eliminate.

JJK

Spot|DSE wrote on 6/29/2004, 4:57 PM

Dang! That's a HUGE difference. How did you end up with such low-level dialog? I'd use volume to punch it up in Forge first, foregoing compression. Normalize in Forge didn't take it to the peak of -3dB? How did you apply it overall? Do you have a peak in there somewhere that is far louder than everything else?
In this instance, get your levels correct first, then apply EQ, then apply Noise Redux. Then apply compression. Depending on the file, you may want/need to apply compression first, then Noise reduction.

musman wrote on 6/29/2004, 6:58 PM

Thanks, Spot, I appreciate it. If you want a description of why the volume was recorded so low, it's below.

I've experimented a lot, but I generally end up:
1- first evening off the clip by lowering the loud parts.
2- Normalize to the 'speech' setting to -.03db
Then things get tricky, as there's a lot of audio noise.
3- EQ the audio noise out as much as possible. I've tried the high shelf hiss removal parametric eq, different paragraphic eq stock settings, and finally a 20 band graphic eq and going through the different frequencies to find and then lower the audio noise. I could swear sometimes the same things work on a clip and sometimes they don't- maybe this is an issue with stacking things in Sound Forge's plug in chainer. A lot of times it sounds different in preview than it does when I process the chain to the clip. That really confuses me.
4-Compression- generally choose maximize volume.
5- Do Noise Reduction. There's often so much to do that I get artifacts and often an echo quality. Sometimes I try to eq the sound back to a more normal sound, but this seems to introduce more of the noise I've been trying to remove.
6- I try to use more compression at this stage and often increase the output as the noise reduction lowers the sound overall a bit.

Am I doing something wrong in trying to do these things all in one audio chain? Again, it often sounds much different in preview than when applied. I use the chain as I hate having to edit the audio destructively, which seems like my other alternative.
Does this workflow sound good, or sound I try a different way? What are the advantages of using compression first then noise reduction instead of the other way around?
Damn, this is a mess. But if anyone can help me out I do appreciate it very much.

-----------------------------Crappy Dialog Description------------------------------

The film with the lousy audio was made in one of those make a movie in 48 hour contests and I didn't have many people to help out, so my idea was to make a reality show (about the last man in America who's never been on a reality tv show) so I could double up cast and crew. That way, I figured, I could get the boom mic guy in close and wouldn't have to hide him. Well, we still managed to screw it up and recorded into a Sony pd150 (which has lousy sound to begin with and still has a hiss even after Sony 'fixed' it in later the later serial #s). In most places the recording is way too low but in a few places it's too hot. I'm thinking part of the problem is the mic (Sennheiser me66) is very directional and often is not pointed at the speaker (like when the boom mic guy says his lines).
Add to this a the necessity of very run and gun type atmosphere (both to follow the quick action and improv of th actors and limited time we had to shoot) and also poor monitoring and there's a ton of problems. Why did we monitor so poorly? 2 reasons: 1) I tried to keep the audio settings at and just below 50% with mic att turned on. This is the suggested settings by Jay Rose at Dv.com's audio forum 2) We monitored with 2 sets of headphones split via a radio shack adaptor from the camera, so both headphones were too low to get a good idea. It sounded okay there and then, but when pumped up to -3db its short comings are numberous and profound.

Here's what I'm wondering:

1- What level should dialog be at on SF's meters? Think I was told it should peak at around -3db and I'm guessing you want to keep the rest of the dialog as close to this as possible. Is this right?

2- What is the best way to get my audio uniformly near the -3db? What I've tried so far is to:
a) go pick out the occasional loud spots and lower them with the volume tool.
b) Then I use the speech preset in the "normalize". This brings things up a lot.
c) Finally I use the Maximize volume preset in Compression. This brings things up close to or over the -3db.

3) The way I've done things as described above results at this point in the noise inherent in the lousy recording being very noticeable. To remove the hiss, etc, I first use the 'hiss removal' (a high frequency shelf) preset in parametric eq. Then I use the noise reduction and play with that a bit.
So, is there a better way to do this? Seems like I'm using the same kind of tool in the Normalizer and the Compression twice and not sure if this is a good idea. Also, should I use the Noise reduction and hiss removal tools earlier in the process? Though it might make a difference in what the Normalizer and the Compression use as reference for their most quiet sounds.
Sorry for the long post, but I'm a newbie with sound and am trying to learn the best ways to use Sound Forge. Any help is appreciated!