Homemade Exponential Horn Project

[Ethan]

Fiberglass does sound rather nasty—didn’t realise it was quite that bad. Sanding by hand doesn’t sound too bad to me, but then I did sand the entire former for my horn by hand at least four times (one to get the shape right, two to smooth the surface, and one to polish the shellac), so my perspective may be a little skewed there…

To judge only from one record didn't give the complete image except for that particular record.

That’s a good point—mostly, I’ve tried looking at the responses of multiple records individually, or before and after adding layers to the horn, to see if the response changed any. Combining multiple recordings together before looking at the plot spectrum is a good idea, though—I’ll have to try it sometime.

If possible, I would definitely try to use good late-‘20s or early-‘30s electrical records for testing, instead of post 1935-ish; I think companies started switching over to constant-amplitude recording sometime in the mid-to-late ‘30s, so later records would tend to have too little bass and too much treble.

I once used Audacity to look at the frequency content of a modern military band recording of Scott Joplin rag, and it had that same decreasing response in the treble—Could it just be that in real music—as opposed to test records—the average levels of higher frequencies just do decrease as frequency increases?

[anchorman]

I managed to snag some test records a little while back, from a very nice man in Connecticut. I’ve been worried about tearing them up with steel needles, so haven’t used them until today, since I recently got a sharpener for Burmese Colour Needles fixed up to be usable again.

Attached are two spectra, first from an HMV 5B soundbox, and second from a Columbia No. 9 for which I made an adapter to fit the HMV 102 tonearm.

Interestingly, I could distinctly hear the 10khz tone being played on the record, but it was not much above the noise floor. Things got progressively better from there. The HMV has a distinct peak at 4khz, and then drops again until close to 2khz the response evens out. The Columbia was able to hit the 150 hz note much louder than the HMV. I also noticed, that the harmonics produced in the higher frequencies above 4khz tended to favor harmonics almost as much as the fundamental. In some instances, such as the HMV5b at 6khz, the second harmonic was significantly higher than the fundamental. This would explain why the HMV sounds “brighter” than the Columbia soundbox, and is probably an artifact of the spider used to attach the needle bar to the diaphragm.

Both cleaned up and were very respectable looking in terms of harmonics between the 3-4khz range on down to 200hz.

The overall surface noise was greater in the 200-4000 range, such that it almost dominates the higher frequencies, but when a higher frequency tone is played by itself, it is very clearly heard, despite the broadband noise that is louder in the mid and lower frequencies.

All this to say, see if you can snag some test records, there is interesting information to be had.

Looking forward to trying this out with a microphone better than what is on my iPhone, and some slightly easier to use software. I am still a rank amateur regarding windowing and other aspects of FFT capture, but this tool does give usable information. Even more so, looking forward to seeing the results with a great big horn instead of this little portable.


The HMV 5B:

HMV 5b spectrum

The Columbia No 9:

Columbia No 9. Spectrum

And the test record used:

[anchorman]

Any of the info above 10khz in the above pictures is harmonics… the test record starts with 10khz tone after the initial 1khz tone to set the level.

[anchorman]

My experience has been as Ethan pointed out, that the bulk of the info on these records is in the 200-2000 range, trending towards the lower side. This makes sense, since the human voice range would dominate most recordings where there is singing involved.

I haven’t tried yet a comparison between new and old electrically recorded discs to see how things may have shifted.

[Ethan]

Interesting! I'm a little unclear, though--is that record a sweep, or individual notes? If it's a sweep, it's odd that both soundboxes seem to drop off somewhat below 5000 Hz, as that's what seems to be the accepted maximum frequency for matched-impedance soundboxes. It will be very interesting to see what the response is on a bigger horn--and how the response differs on a machine with a lid, as that would cut out some of the treble, compared to a lidless portable like the 102.

[anchorman]

Individual notes - 10khz, 9, 8, 7, etc. it’s listed on the label.

There are sweeps on some of the other records that I got, but I was not able to change the smoothing of the curve in the spectrum analyzer easily enough with the program I was using, so I decided to try this other recording. The interesting thing with individual tones is you can look at harmonic content if you just do the single tone. I’d like to figure out some sort of mechanical driver for the needle bar so that one could run such tests without the test records too, but at the moment that’s another project for another lifetime!

It would definitely be interesting to work out what kind of driver works best to connect directly to the horn to look at just the horn’s frequency response. I saw you had done that with a small headphone driver, earlier. I’m not sure yet how loading the driver with the horn effects the driver’s frequency response.

[Inigo]

All those experiments seem very interesting. But for the aim of looking at how a specific machine reproduces sound from records, the simple experiment is what you need. Just play records in succession while recording in a continuous file, and then extracting the spectrum. Of course all the machine resonances will be there, but that's what is interesting to analyze, even the resonances of the room... All factors together produce the sound you're listening to.

[anchorman]

The problem with that is that it doesn't allow you to isolate any one aspect of reproduction from the others in order to try to correct that problem most effectively. Sure, we can determine that this combination of parts that is such and such machine makes these sounds. What was interesting earlier was putting the Columbia soundbox on the HMV machine and getting the "columbia sound" out of the HMV machine. Easy experiment, and now that I've seen the spectrum of the columbia soundbox, I can understand little bit more why it has the "columbia sound".

[anchorman]

The nice thing with the continuous tones is that it more easily allows you to see what the harmonics are and what the fundamentals are actually being reproduced. Taking the whole spectrum of pink/white noise, or taking the whole spectrum of an entire song doesn't isolate out the harmonics. Playback, of course doesn't isolate the harmonics either, but it is interesting to see what is what. At some point I'll need to play this record back on a high quality modern system, and see what the level of the harmonics in the recording itself are.

[Ethan]

I’d like to figure out some sort of mechanical driver for the needle bar so that one could run such tests without the test records too

I think there was something like that in the ‘20s or ‘30s—something where the needle rested on a vibrating element instead of a record, so that a radio could be played through a phonograph horn—don’t know if there are any still around, or how well they worked, though.

I can’t say either about the horn affecting the speaker’s response—I tried measuring its response without the horn, playing right into my computer’s microphone, and it looked as though the same steady decrease starting around 200 Hz was present as when I played it through the horn, except that it had a slight peak around 185, whereas through the horn, it had a trough beginning a little above 130 and ending around 200. There may have been a difference at higher frequencies, too; I was only looking at lower frequencies at the time.

It would be interesting to see how the response of large-horn gramophone (or even a smaller one) changes in an actual room as opposed to an anechoic chamber—it seems as though it could be quite different, depending on the room size and characteristics.