Help cleaning up recorded phone conversations

[sredmyer]

As I posted in an earlier message I have several hundred micro cassettes of recorded phone conversations I am recording to PC for archive purposes. These tapes were made by many different people over several years time with different equipment. As a result some of the recordings are very good while others are very bad. The biggest problems are:
1) One side of the conversation is badly distorted (like the person was speaking to close to the phones mouth piece)
2) One (or both) side very quiet barely audible
3) Excessive background noise (mostly typical noisy recorder but other background noise creates problems too)

The question is, are there any prebuilt filters that can be applied in batch (chains) mode that would produce satisfactory results? I am not looking to turn lead into to gold here just hoping that I can improve the quality on the worst to make them listenable while at the same time not hurting recordings that are already ok.

I have tried running the Normalize, Leveller and Noise Removal effects but on whole I would have to say the results were not much help (especially on the quiet recordings). I suspect the poor results have more to do with my ignorance about how to best use the tools than the ability of the tools though.

Can anyone suggest what filters/effects I should use and what settings would have the greatest positive impact for the problems mentioned?

Thanks in advance for any help,
Steve

[sredmyer]

I thought maybe a more complete explanation of what I am doing might help to generate the most useful solutions. So here is that explanation.

I am recording these phone conversations (currently saved on micro cassettes) to wav files on my PC. The playback of the tape, for recording to the PC, is done at high speed. I do this in batches, recording about 100 tapes at a sitting. Then I run the Audacity Speed Change effect to slow the audio down to normal speed and convert the wav to mp3. I do this in batch mode (Chains) overnight.

As mentioned in the previous posting I have tried fixing the audio problems with effects/filters applied as part of the chain during the overnight conversion.

The actual chain I have tried is:
ChangeSpeed: Percentage=-50.000000
Normalize: ApplyGain=yes RemoveDcOffset=yes Level=0.000000
NoiseRemoval:
Leveller: dB=-70dB Passes=2
Normalize: ApplyGain=yes RemoveDcOffset=yes Level=0.000000
ExportMp3:

Which was adapted from the "CleanSpeach" chain example.

While the speed changing bit works a charm, the rest does not do what I would expect. There is minimal improvement to the background noise (probably to be expected since there is no concrete definition of "noise" for the filter to work with) and in general the entire recodring is made louder but that is all I can tell is happening. What is not happening (that I really need) is for both sides of the conversation to be brought to a similar level. Since the entire recording is amplified by the application of the above chain it does help in the case where one side of the conversation is nearly inaudible, however, there are still huge differences between the two sides. Also I would expect that once run through this processing all files would be of a similar level...that simply does not happen. Quiet (difficult to hear) files are still quiet and difficult to hear after processing.

Again I do not think it is the tools that are the problem but rather my ignorance about how to apply them. I am hopeful that someone here can help me to understand the parameters for the filters (or maybe I just need different filters) so that I can achieve my goal.

Thanks,
Steve

[kozikowski]

We're clear what you want to do. The reason nobody's leapt to your rescue is there is no rescue.

We would be hard put to fix these performances one at a time by hand and some of them are in doubt. Chains will only manage simple, predetermined tasks. So, no, I don't think this job is going to happen so fast -- at least not in Audacity. You're talking about law-enforcement forensics tools.

Koz

[Irish]

What you are doing doesn't really lend itself to batch processing.

The biggest problems are:
1) One side of the conversation is badly distorted (like the person was speaking to close to the phones mouth piece)
2) One (or both) side very quiet barely audible
3) Excessive background noise (mostly typical noisy recorder but other background noise creates problems too)

Each of these problems requires a different solution, and you shouldn't be processing audio that doesn't need it.

ClipFix may help with problem 1), but you have to apply it only to the affected part of the track.
Amplify will help problem 2), but it will also amplify whatever noise is there as well.
Noise Removal will help with problem 3) and with the amplified conversations from problem 2) but it will not do anything without a noise profile.

There is minimal improvement to the background noise (probably to be expected since there is no concrete definition of "noise" for the filter to work with)

You said it yourself!

This page from the manual
http://manual.audacityteam.org/index.ph ... se_Removal
describes how the Noise Removal tool works, but it has to get the noise profile for each tape, because each will be different. I don't know any way to do that other than manually.

PO'L

[sredmyer]

We're clear what you want to do. The reason nobody's leapt to your rescue is there is no rescue.

We would be hard put to fix these performances one at a time by hand and some of them are in doubt. Chains will only manage simple, predetermined tasks. So, no, I don't think this job is going to happen so fast -- at least not in Audacity. You're talking about law-enforcement forensics tools.

Koz

I realize that I will not be able to perform miracles but I was hoping to at least improve on the situation. Specificaly I was hoping to fix the volume level issues both overal volume (when entire recording is exteremly quiet) as well as the problem when only one side of the conversation is quiet. From my reading regarding the Normalize and Leveller effects this is exactly the types of problems these effects should help. However I am not seeing any improvement (on these issues) when using these effects. This is why I suspect it is the parameters of the effects not being set appropriately that is causing the less than noticable results.

Am I wrong in my understanding that this issue can be helped with these tools?

Thanks,
Steve

[sredmyer]

This page from the manual
http://manual.audacityteam.org/index.ph ... se_Removal
describes how the Noise Removal tool works, but it has to get the noise profile for each tape, because each will be different. I don't know any way to do that other than manually.

PO'L

Yes I read that page but when I saw that the Noise Removal tool was applied in the example CleanSpeach chain I though maybe there were some default noise profile(s) which were used when applied in batch. So is there some way to create a "default" profile for specific noises (ie. tape hiss, 60 cycle equipment hum, etc.) that can be applied to files in a batch?

Thanks,
Steve

[kozikowski]

Step one of Noise Reduction is to get the noise profile. After that, it keeps the same profile unless you restart the program or get a new profile. I don't know of any default value.

The Program Sound Soap has default values. You tell it which kind of utility power you have and it automatically looks for hum at that frequency. But the developers are really clear that automatic correction is a faint shadow of what you can do by capturing a profile.

Chris's Compressor will level out variations in volume and I use it all the time on perfectly recorded shows. It doesn't do well with noise because the background noise starts pumping and throbbing in the finished show.

http://pdf23ds.net/software/dynamic-compressor/

You're holding the Perfect Storm show. All the seriously bad errors in one place -- or many hundred places.

Koz