Enhancement problem

[O'Manannan]

I?m trying out Millennium on the 15 day trial. It works well for cleaning noise from music, but I also need to enhance a telephone conversation recording made with a cheap cassette recorder on a plain vanilla tape. The record level is very low and the noise level is high. Turned up loud you can make out most of what the local end (male voice) says, but only about 20% of what the remote end (female voice) says. I?ve tried all the ?tips and tricks? for telephone recorgings but none do much good individually, and with 6 or 7 applicable filters/effects/forensics the number of combinations, and the orders in which to apply them, verge on the infinite. The need for this is not nearly as exciting as one might guess, but it is important. I don?t do detective or police work, so I can?t spend $1,000.00+ on forensic software I?ll use once. Anyone have any suggestions?

[DougMac]

Chicken Salad

Before going any further, I'd suggest taking a section of the file with the female voice and doing a test restoration. This will eliminate the difference in amplification from interfering with the restoration.

Run a CNF by sampling in a quiet portion of her conversation. Then do a gain normalization followed by more cleanup work. If you can get acceptable results on the test audio, you can then go back and do a full restoration. The trick will be to restore the male and female voices separately, since the signal to noise ratio in the different sections is great.

It could be that the signal of the female voice is so deep in the mud that you'll never be able to pull it out. DC Art is wonderful software and I've gotten some remarkable results, but I've found that there are limits. It can't put back what wasn't there to begin with. I'm reminded of TV shows that deal with forensics. They routinely show impossible restorations of photographs and audio.

As my late friend was wont to say, "You cant' make chicken salad out of..."

Doug

[Craig Maier]

After you perform the steps that Doug suggested, you might try applying the "Speech Filter" preset found in the bandpass filter. That may also help a bit. The preset is set for 12 dB/Octave. You may also want to experiment with the same preset but set for 18 dB/Octave.

[Dan McDonald]

Another thing that I've found in other situations (where there's a big difference between parts of a conversation) is to chop the file into pieces, then concatenate the different parts so that you have one end of the conversation in one file and the other end in a different file. Then you restore each one separately, cut them again and put them back together in the proper sequence. I don't think it'd go well for legal (courtroom), but it probably would if both the unrestored and restored versions were played back.

Anyway, this has been helpful for me sometimes.

Dan

[O'Manannan]

Thanks!

I don't think I've ever gotten so much useful info so quickly from any board on any subject!