Subject:Editing and restoring MP3 audio files

Posted by: plasmavideo

Date:6/13/2005 7:05:59 AM

I would like some advice on the best way to accomplish a restoration project.

I need to cleanup and try to restore some old radio shows from the '30s and '40s that were digitized and converted to mp3, unfortunately at 32kbs, 22K audio. Some of them were properly encoded and do not sound too bad - others are pretty bad, mostly due to poor encoding techniques or from someone trying to aggressively apply some poor noise reduction techniques. Others are encoded fine, but have lots of typical scratch, hiss and surface noise and poor fidelity.

Anyway, I know what I need to do to fix most of them using SF and NR2, but here is my question. When I open them in Sound Forge 7, are they automatically upconverted to a higher quality for direct editing and noise reduction, or should I resave them in a higher (44K - 48K) WAV file first and then work on them before re-encoding to mp3 (I need to return them to a higher quality mp3 or wma format eventually). Should I apply any type of noise shaping or dither, and if so, what step in the process? I didn't know if any kind of upconvert/downconvert/re-encode process would add yet too many other artifacts. So far, just opening up one, editing it and resaving in the same mp3 format worked OK, but I want to maximize my chances for success, as I have a number of them to do. Also, if I want to resave at a higher mp3 bitrate should I apply dither?

Thanks in advance.

Tom

Subject:RE: Editing and restoring MP3 audio files

Reply by: Chienworks

Date:6/13/2005 7:38:39 AM

I'll tackle a few of the points, but i'm not an expert in this stuff.

Sound Forge will open the MP3 file in your current uncompressed WAV template format, usually 44.1K 16 bit. When you save it it will be compressed again, even if you save it back to an identical MP3 format. The lesson here is to save MP3 compression for the very last step.

Note that even though Sound Forge will decompress the files upon opening, it does not convert them to that quality. They will still sound exactly the same as the original MP3 files. Sound Forge adds enough bits to fill out the uncompressed format, but it cannot regain lost quality. I think you understand that based on what you've said, but i wanted to say it directly just to be sure. For that matter, if you open a 22KHz file as 48KHz then Sound Forge has to resample it. Any resampling or other processing can and probably will degrade the signal. If you know the files are 22KHz you may want to make sure you open them as 22KHz by setting that as your default template first. Even 22.050KHz to 44.100KHz involves resampling, though it's probably almost completely accurate.

When resaving to MP3 there will always be a quality loss. Even saving to 256Kbps will result in less quality than your original 32Kbps files. Once the file is compressed, not only can you never retain the lost quality, but any further compression loses more quality. Of course, resaving to 256Kbps will result in less loss than resaving to 32Kbps again, but probably not noticeably so. You may want to do some blind comparison tests and see if saving at a lower bitrate loses noticeably more than higher bitrates. Saving at a higher bit rate may not sound usably better than saving at the original rate.

Dithering only applies to bit-depth changes and has no effect on sample rate changes or bit rate changes. Anti-aliasing may be helpful if you change the sample rate. Neither applies to bitrate changes. Also, increasing the bit depth, sample rate, or bit rate will not increase the quality of the file. Once any of these have been lowered, you can't restore what was lost. The only reason to increase bit depth or sample rate would be to get more accurate processing of noise reduction, EQ, or other effects. This accuracy may be lost anyway if you are saving to another highly compressed file at the end.

In general, 32Kbps probably sucks. You've probably lost more to this compression rate than you can gain back by noise reduction or EQ correction. Whatever fixing you attempt probably won't make it sound better enough to be worth going through another compression step and the additional loss that comes with it.

Subject:RE: Editing and restoring MP3 audio files

Reply by: plasmavideo

Date:6/13/2005 9:18:05 AM

Thanks Kelly.

Yes, I realize that upsampling to 44K won't add any quality, and you verified what I was thinking - that upsampling might help when it comes to applying the noise reduction.

Thanks for the tip on dither. I thought it also applied to resampling, not just bit depth. Got some education on that one!

I like your tip for setting up the project template first to 22K. That will give me a bit more accuracy on opening, I suspect, and give me more leeway as to what to do afterwards.

The only reason they will eventually will be resaved in mp3 is for playback on portable players. I do plan on keeping better copies for myself, but as you say, I ain't gonna add more hertz than what is already there!

So far, I've had pretty good results using the vinyl restoration tool and/or NR2 plug-in on several of the files. My workflow at this point has been to first eq to get rid of the hiss and LF rumble, apply the noise reduction or vinyl restoration, re-eq and apply a bit of HF stimulation using the HFX plug-in. I also use a light touch of Wave Hammer to even out the sound a bit. Sometimes the audience reactions are much louder than the talent, due to the early recording techniques. Some of the recordings have turned out remarkably better, but some have the whistle and tinkle artifacts of poor encoding or NR, and they are about hopeless! A shame, as some of these old radio shows (Jack Benny, etc.) are priceless. I've had a lot of fun listening to them as I go. I hope some of these exist in better condition elsewhere, but most of the ones we've found so far are copies of the poor copies.

In addition, some of them run horribly off speed, either from the original source or the encode process. SF has been invaluable getting the speed and pitch correct.

I sure wish I had the original vinyl, ET or tapes to work from. I think that 32kbps was originally chosen so collectors could cram a billion files on one mp3 CD. Whatever the reason, that seems to be the standard found on "collection" CDs or available over the internet. From my research, I've found some individual sources, or places like Radio Spirits that have a number of shows in great condition, but some of the collectors have told me that some things just are not available anymore except in the cruddy 32K format, as original sources no longer exist. Collecting these shows looks like a fascinating hobby. Maybe something to do in the future after retirement.

Thanks again.

Tom

Subject:RE: Editing and restoring MP3 audio files

Reply by: Chienworks

Date:6/13/2005 12:35:29 PM

It is fascinating to get ahold of some of that old stuff. I bought a collection of old radio shows on cassette a few years back. They were manufactured by a company named "Metacom" and i found them in a 'dollar' store for something like 5 for a dollar. The problem is that they used the absolute cheapest tapes they could find. Aside from being afraid that they'll break if i fast forward or rewind them, they don't have slipsheets so the tape vibrates as it plays causing a high-pitched squeal that has actually been recorded onto the tape. I did contact Metacom to inquire about their catalog. They still had a large stock, but at $10 each i wasn't going to bite. The sad part is that if they had spent 25 cents more on a better quality cassette i would have been happy to pay a buck more to purchase.

For pure voice only, i've had good success with bitrates as low as 24Kbps. Add any music, sound effects, background noise, multiple voices, or anything extra though and it really needs 48Kbps or better.