Please advise: Remove background noise on old...

lifelyrics

Training wheels on
I'm converting old cassette tapes to digital, and trying to remove as much of the background noise without distorting the speaking voice. These are 20+ year old recordings of church sermons that we are trying to preserve. Hubby has asked me to try to remove as much of the "shhhh" that is probably pretty customary with cassette tape recordings that are that old, and I have played around a bit with compression and the lo/hi pass filters, but don't seem to be getting the results I was hoping for.

Can anyone direct me to some techniques I can try to help in this process?
 
Thank you, Manslick, for your suggestion. Yes, I am transferring tapes to digital files, and importing on my laptop. I was working with the files in Tracktion 3 (Mackie), but wasn't sure if there was another "filter" (called effect, or send in other programs, I think) I should be looking at/trying. I'll research what reaFIR is, and maybe that will help me figure out what it would be called in my DAW. Otherwise, I may end up downloading Reaper. A lot of you guys use it, so it probably wouldn't hurt for me to get to know it for situations where I'm needing help like this!
 
As you've discovered tape hiss tends to be broad band smear (wide frequency range), thus difficult to 'notch' out, remove with frequency filters. If amplitude is high enough to be generally irritating (interferes not merely with aesthetic but intelligibility) then it's amplitude is high enough to make compression problematic. As tape hiss is spread, more or less evenly, through out program gates might reduce it in contentless portions but that make it's presence in content portions even more irritating.

The task you are describing falls under the general heading of 'restoration' (restoring audio) and there are specialized apps for this type of process. If you have a lot of tapes to work with picking up something like Adobe Audition (a very good editing program (with multitrack support) and very good reputation in the restoration community) or Izotope RX might be worth reviewing. (standard disclaimer: have no affiliation with either product). In point of fact most of the restorers with whom I'm acquainted use both (on same project). I try to avoid restoration projects but I also use both when compelled (to restore).

With a lot time, lot of practice, patience, obsessive attention to minutia there is some stuff you can do with standard features of general audio processing apps (I use reaper for tracking but have not explored most of it's processing/editing features . . . I have a lot of respect for Reaper development team but tools I use are basee on way in which I work, not using Reaper options (for editing) merely says that I have yet to have time to investigate them so don't have opinion one way or the other and find the tools I do have to be effective and efficient. Unlike what some procedural shows on TV suggest an audio file is not made up of discrete elements. That is there is not Hiss AND Conversation there is just Hissconversation. Anything you do to reduce perception of hiss will effect conversation as well (standard metaphor is to reference something like a loaf of bread, once bread is baked a bit difficult to extract discreet components: floor, sugar, water, salt, shortening, yeast, etc.) Generally speaking the smaller the chucks of audio you seek to examine and work on the more successful you'll be. Apply a (frequency) filter across the entire file and you might notice very little impact (or even negative impact), narrow the 'window' of the search, locate those sections on which the filter does display improvement and by the time you are done (if you are not insane) you might have improved intelligibility of entire performance. Same thing is true for compression.

An additional 'trick' you can try is to locate a segment of 'just' the noise you want to remove, copy it, invert it then paste it (section by section with precise physical alignment) across the entire file. Some of the 'noise reduction' plug ins, process work on a similar principle but trade some of the tedium for less precise results.

Depending on quality of material (basically ratio of content to noise) it's quite possible that (from a single 'noise only' section) a single pass by Adobe Audition's Noise Reduction process would be sufficient, 20 sec. To 1 min. Per 4 min. Section of material (more thorough approaches can take quite a bit longer but relating time to 'good enough') Adobe also has separate processes for hiss and clicks as well as an 'adaptive' noise reduction that, like all noise reduction algorithms, works best of material that needs it least but doesn't stink (it tends to be CPU intensive and on 3 gHz dual core systems with at least 2 gig of RAM still takes 2.69 sec per sec of audio to run, and typically requires a couple of passes to 'dial in' what parameters work 'best' on that specific audio)

All of these tools entail trade offs . . . A noise reduction 'profile' that run at 100% produces artifacts nearly as unpleasant as unprocessed file might be the time/cost benefit choice if run @ 33%. The same 'profile' that worked wonderfully on material from tape 'a' might be disastrous on tape 'b' material.

There is a web site called 'Pristine Audio' that presents, examples from restoration of vintage 78's, from 70 yr. Ago (some older some newer) what obsessive attention to minute detail can accomplish. For some hints and points to things that might help you could do an internet search looking for 'younglove approach audio cleaning' (or restoration, etc.) The 'younglove' approach dealt specifically with vinyl but might provide some additional pointers that would apply generally.

Good luck
 
Wow, Oretez, thank you so much for your time, and your great feedback.

I have experienced what you were talking about in trying to fix an extremely old recording of a friend singing a capella when she was young. She actually was in her 70's, and couldn't sing any more, but had a beautiful singing voice when she was young, and came across an old tape where she was singing. She wanted me to make a cd for some of her family of the two songs, but, man, there was a terrible back noise (worse than what I'm dealing with now, probably because the original was so much older). At the time all I had was audacity, and I tried several things to remove the noise, but it made her voice sound electronic and awful. So, I pretty much left it as is (the adjustments were so minor, you hardly knew it had been worked on, unless you heard the original!) Too bad all I have left is the mp3's I created (which means I'm left with a lot less to work with than a cassette tape), and she has since passed away.

So, I'll check into the things you mentioned. By the way, since you've worked with this stuff, can you tell me why the highs and lows are cut off on the was file? I noticed that, even though the sound is not clipping at all, when I imported the files, they look like the waveforms were chopped off of the top and bottom! Weird. I'm going to look into playing the tape directly into our old laptop, into Reaper, I think, and see if it still does that. (I had to use my DP004 to convert to digital, then import to my laptop, because my 1/8" mic-in is broken on my new laptop!) Hey, I just thought of another way to test it, with my firewire interface! OK, I'll be back later to give my results, in case anyone else is attempting to do "audio restoration", and cares to find out!
 
By the way, since you've worked with this stuff, can you tell me why the highs and lows are cut off on the was file? I noticed that, even though the sound is not clipping at all, when I imported the files, they look like the waveforms were chopped off of the top and bottom! Weird.

Addressing the waveform appearance

Without seeing it, hard to say for sure

Was not entirely sure whether you were referring to latest transfers from tape or the earlier musical performance and even speculation might be different depending

Included a couple pictures that might or might not point to possibilities

The first one is essentially a pic of a wav of a live vocal guitar performance without any added FX more or less normal dynamics. While this represents only about 30 sec. It's resolution is roughly consistent with what would let me view the entire 3 min. File on my video display. If you zoom in, or out, depending on intent the appearance of the file will change. The 2nd picture is of the same resolution but the audio has been, for me, hyper compressed (I've worked with people who have been know to compress things even more, whose goal would be to make the audio representation look like a brick). At 'normal' visual resolution the top of the file looks flat.

Third pic has two images. Same Audio this time a small sample size. The Audio amplitude was increased beyond the level that could be represented by 16 bit audio you can see waves on the left (of each image) terminate in more or less typical rounded, pointed (sine wave) fashion. As the actual audio exceeds 16 bit 0 dB the wave didn't reach it's peak amplitude before it was truncated . . . The visual is missing the top portion of each wave . . . Flattened, clipped

What the top of the wave form looks like can depend on relationship of visual representation to audio data . . . Higher resolution reveals more detail about the pressure wave, lower resolution visual representation will look more like a simple geometric shape, diamond, rectangle, etc.

These plots are amplitude over time (only one of several ways to visual represent audio, although perhaps the most common, but do not reveal all information. You can plot frequency over time and frequency over amplitude (and I suppose amplitude over frequency though I don't typically use that one) The peaks (top and bottom) are peak amplitude, at various visual representations the 'dynamics', the less compressed the piece the more variety there will be in those peaks, more like mountains, less like northern Kansas. If you are viewing an amplitude over time plot of one hour, it's top will look flatter then same material @ 1 min resolution Very generally speaking a spoken word piece would tend to have less dynamic variety then, for sake of argument, most music, hence more regularity at any resolution for the top (or bottom) of the wave. If by chance the new content on which you are working was intended for broadcast it would be very common to run the material through a compressor. The job of that tool is to even out amplitude; quash high levels then via make up gain increase the lower levels making a visual representation look more like image #2.

Additionally the potential dynamic variation is controlled by the hardware, microphones, amplifiers, recorders, etc. Can all limit potential variety of a visual plot of amplitude over time. A cassette recorder, as a rule, can preserve less variety (has lower headroom) in an audio pressure wave then can any but the cheapest of current analog to digital converters
 

Attachments

  • wav01 raw.jpg
    wav01 raw.jpg
    58.5 KB · Views: 346
  • wav02 cmprsd.jpg
    wav02 cmprsd.jpg
    59.9 KB · Views: 347
  • wav03 clipped.jpg
    wav03 clipped.jpg
    62.2 KB · Views: 344
Last edited:
Oh, this is sooo cool. Thank you so much for taking the time to help me with this!

First off, how do you get the picture of what is in your DAW into your post? I wanted to do that on my last post, to show you what I was talking about, but didn't know how.

Secondly, you answered, at least part of, my question with your explanation about what happens to the sound beyond 16 bits (the DP004 is a 16 bit digital recorder). I'm sure that is what caused the wav file to end up looking like the 2nd track of the compressed version, albeit there are more variations in the wave form, and many spots where it thins out quite a bit in comparison to where it clips out.

These tapes (I've done about 8 - made mp3's, and have 15 to go) were, simply, someone in a pretty small church, maybe 40' x 50', brought in a tape recorder, and recorded a preacher, from somewhere near the front. No microphone, no technical stuff, but the speaker is decently audible. There is the "room" noise, sometimes you can hear cars or firetrucks going by, which will probably have to stay, and there's the normal tape recorder noise (which is what I was hoping to diminish). Both of the preachers who are on these tapes have since passed away, so the recordings are precious keepsakes.

I'm trying to play the tape into my TAPCO 4x6 firewire interface, to see the difference, which I'm almost positive there will be... I'll be back later, and if you could give me some instruction on how to post the pic's of what I'm working on, I'll get that out as soon as I can!!

Thanks again, Oretez!

By the way, where (what state) is "Hwy 50"?
 
Simplest way is to hit [PrtScn] (typically on the upper right hand side of keyboard) this copies a picture of the current screen to the windows clipboard. You then open pretty much any graphics editor (i.e. Photoshop) paste ([Crtl]+ [v]) clipboard into a document, size to taste (I used size that is current default for emailing photos and it was a bit larger then needed . . . I think size limit per image is 64k (for the forum) viola

US 50 was one of two early 20th century Transcontinental highway's (Lincoln Hwy. 6 was the other) initially ran from Ocean City MD to SF in (not always sunny) CA. In '70's, for somewhat obscure reasons legislation was passed to terminate Hwy 50 in West Sacramento CA sort of ruining it's 'Coast to Coast' nickname. Attached photo is not from either coast.
 

Attachments

  • Hwy50 sum05 03.JPG
    Hwy50 sum05 03.JPG
    63.1 KB · Views: 337
Wow. Very cool. Thank you for sharing that lovely picture. Now I know why Hwy 50 sounded familiar to me. The other day, I was just looking at a US highway map, and finding which highways went coast to coast (or almost) to see what kind of a road trip it would be to just take one road east! That was the one, I think, that I was mentioning to my Hubby that goes near Lake Tahoe (I love Lake Tahoe), and right by 2 or 3 of the great lakes!
Ok, so, here's a pic of side A and side B of the tape. I didn't adjust the volume; side B is not as clipped off because they are two different people, and the guy on side B had a softer voice. So, doesn't it make sense that it's because the DP is 16 bit? The other problem I have is that whenever I use that digital recorder, then transfer the wav files to a DAW (audacity, Sonar 7XL, or Tracktion 3), they are so tiny! I've asked about it before, somewhere, but now I have a picture to show what I mean!


I tried to test, using my firewire interface, to compare what kind of wav forms I would get, but my laptop crashed twice when I tried to record. We downloaded Reaper last night, and will be trying direct input from my little tape recorder/player through 1/8” out/in to the laptop mic-in.
 

Attachments

  • wow.jpg
    wow.jpg
    64 KB · Views: 337
p.s. Hubby had a theory about why the files are so small from my DP: If I'm recording at 16 bit, and then importing it to 24 bit, it would be exponentially smaller than something that was recorded at 24 bit. So, I'll be testing that theory soon, by importing one of the wav files from the DP into Tracktion at 16 bit (once I figure out how to change that setting!). It's something that has bugged me for some time, and I couldn't figure out what was up. Once he said that today, it seemed so reasonable.
 
p.s. Hubby had a theory about why the files are so small from my DP: If I'm recording at 16 bit, and then importing it to 24 bit, it would be exponentially smaller than something that was recorded at 24 bit. So, I'll be testing that theory soon, by importing one of the wav files from the DP into Tracktion at 16 bit (once I figure out how to change that setting!). It's something that has bugged me for some time, and I couldn't figure out what was up. Once he said that today, it seemed so reasonable.
That wave "height" in digital is directly related to the actual signal volume being recorded or edited, and has nothing directly to do with the conversion from 16-bit to 24-bit.

If your signal is coming in that low off of the tape, either the tape was recorded really quietly, or the volume at the input of your A/D converter (computer interface) is set too quiet.

If the waveforms are larger and of more normal "height" in the 16-bit version - i.e. they are recorded to digital OK, but it's only the 24-bit version where they are that tiny, that's not because of the the word length change itself. A 0dBFS (for example) signal in 16-bit will remain at 0dBFS when converted straight to 24-bit; the "height" of the waveform will not change. But if that's what's happening to you, and if we assume that there's no change in the vertical display scale of your timeline display, then the problem probably is that there's a gain/volume control in whatever you're using to make the conversion that's turned down too low.

G.
 
Hi, I'm sorry but I did not take the time to read all the posts but I might be able to help. I have 3 DBX 150 Noise Reduction units I use with my reel to reel to diminish tape hiss. Type 1 is for reel and cassette. If you think I can help I can run your tapes through my cassette deck with the reduction and see how that works. PM me if you need my help.
 
As one who has done a lot of this stuff, and at the risk of sounding "negative", there is often not a lot you can do to improve a poor recording.

Many of us are familiar with the standard TV forensics crime show scene where they take a really fuzzy, garbled sounding audio tape to the lab, the technician is seen in front of a big monitor screen with complicated looking waveforms on it, and is seen to press a button called "enhance" or 'restore' and everything magically comes back into full clarity.
Same with photos. It's that magical 'enhance" button that does the trick...

There has been one audio restoration software product called something like "sound soap" whose name in a way sums up the misconception that the problem is just a matter of "digitally cleaning" the noise from the program material, like putting your laundry in the Hoover. If only that were possible.

The reality is you cant bring back information that never got onto the recording in the first place. Put another way, you cant unscramble eggs. If there is noise mixed in with the audio you want, and is of the same spectrum, there isnt anything you can really do about it, short of noise gating, which is sort of like shutting the gate after the horse has bolted.

For many years most church sermons for example were probably recorded onto a standard "normal' cassette tape, straight from the desk, with no compression, limiting or noise reduction on the way in. It's a recipe for a hissy and/or distorted recording from the start.

People who work seriously in audio restoration of analog recordings spend most of their efforts on the analog side of things because that's where most gains are to be had. Just having the analog equipment and skill to actually play the tape back at its optimum (even before you do anything to record it to digital) is where it's at. But even then, the gains will probably only be modest as most of the loss of fidelity was probably already done at the initial recording stage.

You have to work with what you have, which often isnt much!

All the best, Tim
 
Last edited:
I have 3 DBX 150 Noise Reduction units I use with my reel to reel to diminish tape hiss. Type 1 is for reel and cassette. If you think I can help I can run your tapes through my cassette deck with the reduction and see how that works. PM me if you need my help.

The dbx NR decode units would be only appropriate (and then they would be almost essential) in this situation if the tapes were dbx encoded at time of initial recording.
As they were church sermon tapes almost certainly made for direct duplication onto other cassette copies for end users that's highly unlikely.

BTW dodgespan, you got it round the wrong way. It's type II that was designed more for cassettes and type I for high speed open reel. Both were companding systems designed to be used at both record and playback stages.

Tim
 
Hi guys!

Thanks again to all for your thoughts, offer to help in my project (Dodgeaspen), advice, and sharing of your knowledge.

I'm so excited about what I've learned from you guys, and what I've come to in this project, and just wanted to share.

First of all, I figured out how to record directly from my little cassette tape recorder/player to my firewire interface, into my Tracktion 3 (Mackie DAW)! In the picture below, you can see that my problem of the high and low frequencies being chopped off has been solved. This is truly an upgrade for me, since I had several steps I used to go through (for the previous 8 tapes), and then I was working in Audacity, which was ok at the time, but is no comparison to what I'm able to do in Tracktion!

Secondly, I played the mp3 from one of the talks I just finished and my Hubby and my family were very impressed with the results! They noted a major improvement in the sound quality. I think you can see that I used some compression and lo/high pass filters on those tracks (wav picture in previous post). I also needed to increase the gain, and use a volume gain in one of the compression parameters, which you can't see from the picture.

I'm so happy, because I feel like I can successfully complete this project in the near future, and it's not going to be a drudgery at all! I'm so grateful to my dear Hubby for his willingness to provide such nice tools for me to use.

See you guys around, in the threads!
 

Attachments

  • Direct to Tracktion.jpg
    Direct to Tracktion.jpg
    37.4 KB · Views: 40
The dbx NR decode units would be only appropriate (and then they would be almost essential) in this situation if the tapes were dbx encoded at time of initial recording.
As they were church sermon tapes almost certainly made for direct duplication onto other cassette copies for end users that's highly unlikely.

BTW dodgespan, you got it round the wrong way. It's type II that was designed more for cassettes and type I for high speed open reel. Both were companding systems designed to be used at both record and playback stages.

Tim

Thanks for pointing the out to me.
 
Back
Top