Frequencies below 20hz present in musical recordings.

[audio_tony]

What makes you doubt them?

Blackman-Harris window is probably better for low frequencies and you need to get the FFT size 16 k or larger to have good resolution at the lower frequencies.

I have tried other software which matches what Audacity shows.

I've loaded files into Audacity and run the spectrum analysis, and then done the same using WaveSpectra (+others) and seen differences.

This was at least 2 years ago now, so I forget the finer details, but IIRC these were single tones generated by Sox.

I may try it again some time and report back. Perhaps I was doing something wrong, I don't know.

 

[Juhazi]

I've loaded files into Audacity and run the spectrum analysis, and then done the same using WaveSpectra (+others) and seen differences.

Most likely this comes just from different default parameters for analysis. Paint eg. same five seconds of a piece, set windowing and FFT size the same.

 

[audio_tony]

Most likely this comes just from different default parameters for analysis. Paint eg. same five seconds of a piece, set windowing and FFT size the same.

I just tried Audacity again, and I remember what the issue is for me. The spectrum plot won't go below -90dB and I can't see any way of changing the scale.

 

[scott wurcer]

This keeps coming up, you can't read DNR directly off an FFT without accounting for sample rate, window, and FFT length. For a full scale sine wave at 16/44.1 and FFT of 65536 the noise floor on a bin to bin basis is more like -138dB

EDIT - I just noticed something strange the old CoolEdit and Audition 1.5 give a slightly different answer for noise floor with exactly the same settings even dither (which I forgot to mention). The dither used will slightly affect the noise floor.

EDIT - Apparently Adobe changed the algorithm for going from linear to log display in the FFT tool, have to take a file over to a real math tool to check.

 

[RayDunzl]

The spectrum plot won't go below -90dB and I can't see any way of changing the scale.

 

[Patrick1958]

I just tried Audacity again, and I remember what the issue is for me. The spectrum plot won't go below -90dB and I can't see any way of changing the scale.

In Audacity top menu edit/Preferences/interface there change the DB (forth line from top) to -145 dB

 

[House de Kris]

View attachment 37581

Hmm... How does it get down to -117dB on a 16bit CD?

Hmmm, sometimes I don't know if people are joking or not on this forum. I'll assume that's a legit question, and bite.

The record length is reported to be 65536 samples. This implies an NIF of 45dB for a very linear ADC that captured this data, or 42dB for an ADC with non-symmetrical quantizations. This would yield a usable dynamic range OF THE DISPLAY of between 138dB and 141dB.

 

[audio_tony]

In Audacity top menu edit/Preferences/interface there change the DB (forth line from top) to -145 dB

Do you know.... I went looking for something similar and found this....

But now you have pointed out the other option....

Although I note that -145dB is not really low enough and additionally I've just noticed the level is wrong (it's -6dB correct under the cursor - but incorrect in the scale to the left which suggests ~ -30dB see screenshot above)

 

[RayDunzl]

Although I note that -145dB is not really low enough and additionally I've just noticed the level is wrong (it's -6dB correct under the cursor - but incorrect in the scale to the left which suggests ~ -30dB see screenshot above)

The level of the graphed data jumps around if you change the width of the display window.

Agree that the Frequency Analysis window is not one of the stronger points of Audacity. It does have other appeal.

 

[audio_tony]

The level of the graphed data jumps around if you change the width of the display window.

And so it does, never noticed that before!

I agree that for other purposes it's very good though.

 

[RayDunzl]

Hmmm, sometimes I don't know if people are joking or not on this forum. I'll assume that's a legit question, and bite.

Legit.

The record length is reported to be 65536 samples. This implies an NIF of 45dB for a very linear ADC that captured this data, or 42dB for an ADC with non-symmetrical quantizations. This would yield a usable dynamic range OF THE DISPLAY of between 138dB and 141dB.

Noise Improvement Factor...

Thank you.

 

[DonH56]

That is true of most piano notes I think.
We have a Steinway Model B and I remember setting up microphones many years ago playing a few single notes to get levels and the overtones were almost as strong as the fundamental, and on sustain the C below middle C had the second harmonic sometimes growing larger than the fundamental as their levels changed.

Not sure about "most" but the lowest octave (remember the lowest fundamental note on a standard 88-key piano is 27 Hz) the overtones (not necessarily harmonic) is often been more readily heard -- see equal-loudness curves. That said, when I have measured, the fundamental is usually the strongest (in absolute SPL) for those lowest notes when struck alone even if we do not hear it as loudly as the overtones.

Unrelated to quoted post: A useful reminder independent of FFT windowing and resolution is that, while the SNR from quantization noise for an N-bit converter goes as 6*N dB, the noise floor (SFDR) goes as 9*N dB. If you take all the little noise samples and integrate them (more or less, the math gets rather involved) you get the 6N number. I am pretty sure I have a little post on ASR explaining it with pictures and all but couldn't find it with a quick look (I need to put together an index like I had on WBF, just haven't had the time or interest lately).

 

[MRC01]

You can use the normalize function in Audacity and one of the check boxes is Remove DC.

True. However, I've found that with Audacity, the DC offset remove doesn't always work as expected. A high pass filter with a low corner frequency like 5 Hz seems to remove DC offset more consistently.

 

[Blumlein 88]

And so it does, never noticed that before!

I agree that for other purposes it's very good though.

What is happening is if you have 32,000 points on a graph displayed on a 1980 pixel wide display you have to average them together somehow. Audacity does a very poor job of this. One of its worst aspects. You can have a tone just a few db from max level disappear. Move the graph around on the desktop and it may reappear. Other software shows the highest bin level in however many bins it has to put together for display purposes. I also dislike that the display only goes down to -145 db. They should drop it to 200 db in my opinion.

If I'm looking for higher level tones, I reduce the FFT to 2048 which will split it up into 1024 bins. This way you get at least one pixel per bin and nothing disappears.

The other nice feature is you can export the data from an FFT graph. This saves it as a text file which can be opened into a spreadsheet. Here it lists each bin with the proper value. This is not very convenient for most purposes, but sometimes it is better than any other way.

The spectrogram fortunately lets you set any range you wish. Just note it also has a gain setting. Something that will push the bottom of the range up fooling you if you aren't careful.

Here is an export of Data from an FFT showing that it will record lower levels than the display. This particular one being white noise at -152 db with a 2048 FFT.

 

[ajawamnet]

the best I've seen out of tape is one of Tim d'Paravichi's 1" 2 track Studer C37s. Custom heads running at 15ips and it extended to below 11 hz. (I would not call it flat, head bumps)

Most 2 track of pop music (that includes rock) was done at 15ips, because it did have a better low end. Jazz and classical was often done at 30ips, they needed the increase in S/N.

I wouldn't trust any of these digital spectrum displays. I trust my HP test equipment, but not something that was thrown on as an afterthought. As a general indicator they are wonderful, accurate, well maybe not so much. Especially at zero as Amir pointed out earlier.

oops, the attribution on Jack's tape curves is wrong, there is no such thing as an A827 2 track, that was only made as a multi-track, 16 or 24. The last 2 track Studer made was the A820.

Alan

That was my fault - I meant 2" not 2TR... That was a 24 track head stack on Jack's site... the 16 track head stack was a bit better but most of them were lost in a fire... I recall some people running them at non-standard speeds to get the benefits of both worlds - 20ips was used from what I recall.

I installed/repaired Studers for years - here's one I did near Pgh - on my site about how bad it hummed in repro (the former A80 did too) due to a blown neutral at the VFW across the street:
http://www.ajawamnet.com/ajawamnet/studiohum.html

 

[FrantzM]

Thanks for sharing. What an interesting story. Electrical systems Grounding is much more involved and complex than most, including power engineers realize, let alone “ mere” electrical engineers

That was my fault - I meant 2" not 2TR... That was a 24 track head stack on Jack's site... the 16 track head stack was a bit better but most of them were lost in a fire... I recall some people running them at non-standard speeds to get the benefits of both worlds - 20ips was used from what I recall.

I installed/repaired Studers for years - here's one I did near Pgh - on my site about how bad it hummed in repro (the former A80 did too) due to a blown neutral at the VFW across the street:
http://www.ajawamnet.com/ajawamnet/studiohum.html

View attachment 37885

 

[audio_tony]

Electrical systems Grounding is much more involved and complex than most

Indeed it is, and when I see people advocating the installation of an additional ground spike for their HiFi it's a recipe for disaster (here in the UK) due to our earthing scheme.

One could easily end up with two earths of differing potentials.

 

[waltzingbear]

Ha! that's very nearly identical to one I had a decade ago. Sparks when a basement ground from front to back of the basement was disconnected, with the service disconnected. Yousa No love for pickups or the Studer A820s

Telco, local power grid co, etc. Never did get the power company to say what they did, but it did go away one day. Later we found the signature of the ground harmonics were of a 6 phase industrial rectifier stack. Somewhere in the local grid. Walking around, the manhole covers and the wate shutoff covers were hot as hell field wise.

Its really telling when you start to peel the onion back and you get to the point that the service is disconnected and the interference fields are still there.


reply to above post:
that is a recipe for disaster any where in the world. ONE GROUND TO RULE THEM ALL

 

[scott wurcer]

Unrelated to quoted post: A useful reminder independent of FFT windowing and resolution is that, while the SNR from quantization noise for an N-bit converter goes as 6*N dB, the noise floor (SFDR) goes as 9*N dB. If you take all the little noise samples and integrate them (more or less, the math gets rather involved) you get the 6N number.

Don't forget that without dither the noise floor spectrum is modulated by input frequency (or input if you are not doing test tones).

 

[scott wurcer]

Its really telling when you start to peel the onion back and you get to the point that the service is disconnected and the interference fields are still there.

You can't seem to win, totally off grid on an island all the folks doing solar have modified sine wave inverters spewing hash everywhere.