Recording volume and data extraction from LP'S

I’ve been going through the forum in the hopes of perfecting my LP recording process. I decided that I was happy using level 7 input volume and then amplifying from there until I came across an LP where there was a pop in a place that was not audible on the record. So I increased the input volume to 9 because I read somewhere that if you record too low, it lowers the amount of data extracted which can make pops sound louder than the music. Is there any truth to this?

There’s a chance that the record had a spec of dust or paper on it the first time I recorded it - no way to tell - the pop was gone with the higher recording volume though and in other areas I didn’t notice crackle or smaller pops nearly as much… Placebo effect?

Now I have read that you should aim for peaks around 0.5 in linear. At volume 9 the peaks were 0.7. I also read that pro audio engineers record between -18 and -12 decibels an the albums where the peaks were in that range sound fine after being amplified maybe even great.

So this next record has pops that show red lines when recorded at full input volume. The actual music is peaking around 0.5 to 0.7. I recorded the album at input volume level 9 and the pops do not go into red but the music is peaking at or just below 0.5.

Am I worrying too much? I am wondering at which point is the recording volume is too low that not enough data is being extracted to get the best sound possible. I’d like to be able to set an input volume that is suitable across the board so that I can record multiple records in a row and export them in batches.

It seems to me that the goal in all of this is to record at an input level loud enough to get the proper amount of data extraction yet low enough that the recording does sound distorted. At which point you use the amplify effect to take it to peak volume. Is this correct?

Thanks in advance

I think you’re overworrying. Peaks at 0.5 or 0.7 are fine, do all your processing and them amplify or normalize ( I amplified my LP and tape collection digitzed recordings up to -2dB and these sound fine on my hi-fi rig).

This set of tutorials should help you:
https://manual.audacityteam.org/man/tutorial_copying_tapes_lps_or_minidiscs_to_cd.html

Particularly this one, a suggested workflow:
https://manual.audacityteam.org/man/sample_workflow_for_lp_digitization.html

Have fun with the project,

Peter.

Not a fair comparison: consumer equipment is 16/24 bit-depth
whereas the Pros record at 32-bit depth, (or more), for the extra headroom.

Audacity pads the file to give it the extra 32 bit headroom

I recorded the album at input volume level 9 and the pops do not go into red but the music is peaking at or just below 0.5.

The problem with loud clicks… the 2nd biggest problem… is that when you Amplify or Normalize it’s using the peaks (the maximum amplitude) as the reference. The music can be amplified more without those “artificial” peaks.

Audacity has a Click Removal effect (automatic) and a Repair effect (manual). [u]Wave Corrector[/u] is a free automatic declicking tool. [u]Wave Repair[/u] ($30 USD) works manually.

For the manual Repair, the [u]Track Drop-down menu[/u] has an option for showing the spectrogram or the spectrogram and waveform together. The spectrogram makes it easier to “find” and zoom-in on the defect (assuming it’s not showing red and already easy to find).

Wave Repair also has a spectrum view. I’ve had Wave Repair for a long time and it can make an audibly perfect repair of most (but not all) clicks. It has several repair options and usually one works. It actually seems to do a better job on bad clicks, maybe because they are easier to find. But it’s VERY time consuming and it can take me a day, or a weekend to clean-up an LP (and it still doesn’t sound as good as a digital original).

Not a fair comparison: consumer equipment is 16/24 bit-depth whereas the Pros record at 32-bit depth, (or more), for the extra headroom.

I believe the pro studio standard is 24-bits/96kHz. Then like Audacity, the mixing & processing is done in floating-point. And since most pro studio recording is multi-track, mixing increases the information, effectively increasing the bit-depth. (Mixing is done by summation* so for example, if you add two 0dB 16-tracks, it requires 17-bits to hold the result.)

-18dB at 16-bits might be “pushing it” but with analog vinyl (and most analog sources) the usable resolution is limited by the analog noise and I wouldn’t worry about -12dB. At 24-bits -18db is fine but there’s usually no reason to record THAT low. If you can’t get it higher low levels can be an indication of some kind of analog problem. Nothing bad happens when you get close to 0dB but you get hard-clipping if you try to go over.

I don’t know where the -18dB pro “tradition” came from but Pro Tools used to work in 24-bit integer so MAYBE it was to leave headroom for effects & mixing.



\

  • In reality it’s more of a weighted average with the individual track levels and the master levels being adjusted for the best sound. Analog mixers (built-around summing amplifiers) and DAWs have a master level control as well as level controls for each input/track. But in floating-point you are still increasing the resolution, downward if not upward.

Audacity has a > Click Removal > effect (automatic) and a > Repair > effect (manual). > [u]Wave Corrector[/u] > is a free automatic declicking tool. > [u]Wave Repair[/u] > ($30 USD) works manually.

For the manual Repair, the > [u]Track Drop-down menu[/u] > has an option for showing the spectrogram or the spectrogram and waveform together. The spectrogram makes it easier to “find” and zoom-in on the defect (assuming it’s not showing red and already easy to find).

Wave Repair also has a spectrum view. I’ve had Wave Repair for a long time and it can make an audibly perfect repair of most (but not all) clicks. It has several repair options and usually one works. It actually seems to do a better job on bad clicks, maybe because they are easier to find. But it’s VERY time consuming and it can take me a day, or a weekend to clean-up an LP (and it still doesn’t sound as good as a digital original).

My goal originally was to backup my LP’S as they are, amplify them and then later on I would fix the clicks and pops and compare to see if the audio quality loss is worth the fix. But I’ve fixed two pops on one record and after playback the music sounds normal. I’m wondering if I should repair the pops from now on and keep it that way as long as it sounds good. But again my hyper-vigilance has me hesitant to do so. I don’t hear loss in audio quality so I must be worrying too much. Am I being silly?

I also have some files that were recorded at half volume; I don’t remember what the peak values were before I amplified. At what level is the volume so low that it won’t properly catch the full dynamic range of the LP?

For my next trick I’m going to try WASAPI because my sound card is 24bit

Thanks for the replies. More reading and re-reading brings up more questions.

Let’s make couple of assumptions - Assuming you are not clipping your analog-to-digital converter (“trying” to go over 0dB), and that you are recording at “CD quality” or better, and assuming no dropouts or “glitches”, the quality mostly depends on the analog-side. (That can include the analog-side of the ADC which can introduce noise.)

I also have some files that were recorded at half volume; I don’t remember what the peak values were before I amplified. At what level is the volume so low that it won’t properly catch the full dynamic range of the LP?

On a “very good day” the noise floor on a record is about -60dB. At 16-bits you have 96dB of dynamic range so it’s rarely an issue.

…Each additional bit represents 6dB so if you are recording at 16-bits and only peaking at -6dB (half volume) you are only using 15-bits. That’s still 90dB of dynamic range, which we would have killed-for in the analog days!

And amplifying digitally doesn’t harm audio quality (assuming you don’t boost into digital clipping).

With digital, loss of resolution is heard as quantization noise. You can hear it if you make an 8-bit file. Like analog noise it’s most noticeable with quiet signals but unlike analog noise it goes-away completely with silence. (If you experiment with this, turn [u]dither[/u] OFF. Dither is added noise that’s supposed to sound better than quantization noise, but dither doesn’t go-away with silence.)

For my next trick I’m going to try WASAPI because my sound card is 24bit

If your hardware supports 24-bits you might as well take advantage of it! I THINK you get the full-bit depth of your hardware by default. Then Audacity automatically converts it to 32-bit floating-point for processing Of course, 24-bit files are 50% bigger than 16-bit files.

Practically, it shouldn’t make any difference… “CD quality” (16/44.1) is generally better than human hearing* whereas analog vinyl is obviously worse than hearing. There is a myth that analog has “infinite resolution” but the analog resolution is noise-limited. VHS tape (analog) is obviously worse than DVD or Blu-Ray and my digital calipers are more accurate than my measuring tape…



\

  • People doing [u]scientific blind ABX listening tests[/u] have demonstrated that a high-resolution original downsampled to 16/44.1 is almost always indistinguishable from the original. In fact, high-quality MP3 (lossy compression) can often sound identical to the original (in blind listening test) or you may have to listen very carefully to identify the difference. (Low-bitrate MP3s can sound pretty bad.)

On a “very good day” the noise floor on a record is about -60dB. At 16-bits you have 96dB of dynamic range so it’s rarely an issue.

…Each additional bit represents 6dB so if you are recording at 16-bits and only peaking at -6dB (half volume) you are only using 15-bits. That’s still 90dB of dynamic range, which we would have killed-for in the analog days!

I’m having trouble wrapping my head around this. I think what you’re saying is that with the bit depths we’re working with that it near impossible to miss out on audio quality - the input volume would have to be SUPER low.

I doubt that at half volume the peaks on that particular record reached -6DB. If WASAPI works well I’m going to start all over again recording with it. In the event that I don’t notice a difference or it doesn’t work; I will record that particular album at a level higher than half volume.

So 32 bit float is for processing. Does that mean that there is no advantage to exporting a wave file as 32 bit float for listening on my hifi system? I’m not worried about space and I have programs that will play the file no problem. But technically speaking signed 24 bit PCM should provide the same sound quality or is the other 24 bit option better?

My plan was to keep a 32 bit WAV file and a 24 bit FLAC file for use on my phone.

I will test WASAPI and see if I can hear a difference. You’re saying that technically speaking it probably won’t matter is this correct?

I tried WASAPI. It works and I thought it sounded better but I tried your 16-bit and 24-bit file save comparison technique and the files are roughly the same size which leads me to believe that my card does not record in 24 bits. It does have 24-bit output. The drivers are up to date for my operating system; I’m not upgrading to Windows 10 to see if a new driver supports a higher bit rate.

I could compare by ear endlessly and not come to a conclusion so I’m probably going to keep using WASAPI and record what I had done with MME over again with WASAPI.

I tried using the bitter extension but cannot for the life of me figure out how it works.

Card recording specs are - 3 Byte 96 kHz Maximum Recording Sampling Rate

I think 24-bit recording should still be possible with my sound card but it doesn’t seem to work that way with audacity.

I tried WASAPI. It works and I thought it sounded better but I tried your 16-bit and 24-bit file save comparison technique and the files are roughly the same size which leads me to believe that my card does not record in 24 bits.

If you export to a different format there will be a difference. Audacity automatically converts everything to 32-bit floating-point so you cant really see what it’s “capturing”. And the drivers can automatically convert the file without telling you… It’s similar on the playback side… You can play a 24-bit 192kHz file on any-old heap 16-bit soundcard.

As you seem to know, there are 8-bits in a byte (24-bits is 3 bytes). For uncompressed audio (i.e. WAV) you can calculate file size as:

(Bit depth/8) x Sample Rate x number of channels. i.e. (16/8) x 44.1K x 2-channels = 176kB per second (about 10MB per minute).

With compressed audio you’ll often see the bitrate (kilo_bits_ per second) and you can divide by 8 to get the file size per second of audio.* With lossy compression the bitrate is a rough indication of quality (more compression = smaller files = worse quality). But you can’t always compare different formats. An AAC file MIGHT sound better than than an MP3 at the same bitrate and FLAC (lossless compression) is about half the size of the original WAV. You can also calculate the bitrate for WAV but with uncompressed files we usually talk about the bid depth and sample rate.

I tried using the bitter extension but cannot for the life of me figure out how it works.

I’ve used it before but it’s not working for me now so maybe it doesn’t work with the current version of Audacity?

Note that any changes (volume changes, etc.) will start using all of the available bits so the file has to be checked before changing anything.


\

  • Except any embedded artwork adds to file size without affecting the bitrate.

P.S.
I think I’ve figured-out how to determine if you have a “true” 24-bit file -

[u]XVI32[/u] is a free hex editor. It’s a programmer’s tool that allows you to see the bytes in any file (in [u]hexadecimal[/u]). You don’t have to “install” it. Just unzip and run the EXE file.

…Hex isn’t a problem because we are looking for zeros and 0-9 are the same in decimal and hex. It’s showing two hex digits (00-FF) which represent 0-255 in decimal. You are looking for “00”.

If only 16 of 24 bits are “used”, every 3rd byte will be zero (the least significant byte or the 8 least-significant bits). There can be other zeros too* but the pattern isn’t too hard to see. Scroll-down to somewhere in the middle of a file.

The trick is to set dither to “none” and export as 24-bit WAV before doing anything.

…I experimented by opening a 16-bit file and exporting as 24-bits. …On the 1st try I forgot to turn-off dither.




\

  • Quiet sounds can have zeros in the most-significant bytes or the two most-significant bytes and and total digital silence will have zeros in all bytes, and once in a while a sample will happen at a wave zero-crossing.

I did some digging on the forum and found that I had to change the settings in windows in order to record 24bits at 96Khz. It works!

I’m wondering if I was to export to 24 bit FLAC, would dithering still be required since audacity would be downsampling from the 32 bit float? Or will it match 24 bits since the recording was done in 24 bits.

I did the test you suggested exporting as 24 bit and 16 bit. The 24 bit file is 50% larger.