Scali wrote:

Here's my Pentium Pro streaming an old track of mine at 44.1 kHz via the CVX-2: https://soundcloud.com/scali/cvx-2-alive-441-khz
Original version is here: https://soundcloud.com/scali/mdb-scali-alive-master-128kbps

The original is 16 bit 44.1Khz, direct digital, and the CVX-2 is 8 bit 44.1kHz, right ?
It sounds great on CVX-2 !
Do you have a soundblaster 1.5/pro / 16 to compare with ? play back this track and record to compare ?

Various Demos/Intros/Cracktros for 286 > 10Mhz (sometimes with Sound/Covox)

dreamblaster wrote:

The original is 16 bit 44.1Khz, direct digital, and the CVX-2 is 8 bit 44.1kHz, right ?

Yes.

dreamblaster wrote:

It sounds great on CVX-2 !

Indeed, you mainly hear some issues in the softer passages. The fade-out at the end doesn't quite want to fade. But I think that is something that may differ somewhat from one machine to the next, if I listen to the earlier 286 vs 486 recordings. It seems the 486 had more 'rumble' in the softer parts.
But I used the exact same setup to record, didn't touch a thing. I only plugged the CVX-2 into the other PC.

dreamblaster wrote:

Do you have a soundblaster 1.5/pro / 16 to compare with ? play back this track and record to compare ?

My 286-20 has a Sound Blaster Pro 2.0 installed. I don't have any software on it to play wav files or anything though. So I'll first have to see what kind of software would play such large files. Then I'd have to convert the song data to that format.

The Covox player was my own little hack, a modified version of the streaming player I wrote for the PC speaker in my 5160: https://www.youtube.com/watch?v=387mfVRLGBQ

Various Demos/Intros/Cracktros for 286 > 10Mhz (sometimes with Sound/Covox)

Wow dr.zeissler, magnificent set of demos !!!

The collection is not ready yet. For each machine I collect the own set. This is my 286Scene for Lowend VGA (>10 Mhz. with SB and/or Covox)
It's really stunning what those lowend-machines are capable of. Thx!

A bit of toppic, but inc201hr (1992) seems to be hires VGA! (but no sound)
Dosbox does not show all effects on this. My Collection works fine on a real machine 😀

If go up to 486/66 there are way more demos/cracktros that will work fine.
So checkout the Hardcode Collection! http://hardcode.untergrund.net (376MB Packed!) Thousands of files!

Off-Toppic => off ! Sorry for the interruption.

shock__ wrote:

The Media Vision "Dongle" is highly complex compared to your average Covox-a-like tho.
Sadly I don't have any pictures of the internals at hand but I think it basically came down to being a ThunderBoard on a LPT port instead of having an ISA interface.

Someday I'll open up my "m-sound" LPT box. It was advertised as being capable of CD-quality stereo audio, but pumping 88KB or 176KB through the parallel-port doesn't seem practical. I am fairly sure there is a DSP in there. The only software I know of that supports it is Links: The Challenge Of Golf (you can see "M-AUDIO" in the Setup section), and a small utility program I was able to grab from Genie in the 90's that reprograms it to act just like a covox so that you can actually use it.

Okay, here's the moment you've all been waiting for: The reference recordings! I received the CVX-2 from Serge a few days ago, set up a testing rig with my E-MU 0404, checked for ground loops, and captured Crystal Dream as well as a test corpus on the following configurations:

Covox Speech Thing (10-pin)
Covox Speech Thing (11-pin)
CVX-2 with no filter
CVX-2 with a 100k resistor
CVX-2 with a resistor and capacitor

You can grab the 24/96 FLAC recordings here: ftp://ftp.oldskool.org/pub/misc/temp/LPTDAC_comparisons/
(You can also play back the test corpus on your own hardware by unzipping "testcorpus.zip", also located there.)
Pictures of the testing: https://goo.gl/photos/evnNxhTj8kh1XE656

I want to stress that nothing changed in terms of my recording equipment between all of these. The recording level was set at "line in" levels and never touched, and I did not perform any post-processing whatsoever. So any differences you hear are differences between the source devices themselves. As for the test corpus, I used lossless original sources where possible (ie. no 112kbps MP3s), and I converted down to 8-bit WITHOUT dithering. Why no dithering? Because 8 bits is not enough resolution to mask the quite audible noise in the upper band added by dithering, and I didn't want that noise to be mistaken for each device's output quality. So you may hear quantization noise in the test corpus, but at least that's easily recognizable as conversion errors.

My impression: The real Covox Speech Things still sound the best. They have a lower output level, but the overall quality is much "brighter" with more "detail". You can also see this by looking at the FFT/frequency displays in an audio editor. Also, the 10-pin and 11-pin speech things sound identical to me, but I did not perform a blind A/B test to confirm this.

How to proceed from here? Maybe a CVX-3 design that copies the FTL schematic, which Great Heirophant suggested was the closest match to the Covox patent?

thanks for the recordings.
test corpus is excellent for comparison.

All sound very clean,
yes the covox sounds brighter, not much bass
the cvx sound a bit muffled, with much more bass (for example at 0:40s in test corpus)

ideally we should get the CVX sound bright, but still have plenty of bass 😀

any ideas ?

I don't think the adapter should "add bass" or color the sound output; that should be left to the user...

What about the resistor ladder/arrangement in the FTL adapter?

of course not, I think CVX misses a lot of bass, it sounds thin, but clear
I compared the FTL ladder with CVX ladder, it is almost identical,
FTL ladder misses one resistor :
on CVX-2 if we short R19, we should get it identical.

you can try by bridging R19 ?
(or replace it by a 0 ohm resistor)

MobyGamer wrote:

Also, the 10-pin and 11-pin speech things sound identical to me, but I did not perform a blind A/B test to confirm this.

Unless my ears are playing tricks on me, the 11-pin version sounds ever so slightly less bright than the 10-pin version.
Perhaps the extra pin is used for some kind of subtle low-pass filter?

MobyGamer wrote:

How to proceed from here? Maybe a CVX-3 design that copies the FTL schematic, which Great Heirophant suggested was the closest match to the Covox patent?

I would be interested in what a Covox clone would sound like with an off-the-shelf R2R ladder pack (of 100k Ohm I suppose, since that seems to be what the real thing uses): http://www.bourns.com/pdfs/r2r.pdf

yes

i compared again and noticed it differs in 1 resistor :
CVX has 1 additional resistor R19, we should test by shorting this resistor
--> see illustration

Hum, that's the last resistor before it goes through the line-out? So all signal would pass through that. That might explain why the 100kOhm version I have is so very soft.

yes i think this is the last piece of the puzzle
with 100k this makes an RC low pass filter
that can fall in the audio range
so yes this could well be the reason for the more muffled sound
please replace that resistor with a 0 ohm resistor
or just bridge it.
i think it will sound good then !

OK, will work on that and reply back with a new set of recordings.

dreamblaster wrote:

yes […]
Show full quote

yes

i compared again and noticed it differs in 1 resistor :
CVX has 1 additional resistor R19, we should test by shorting this resistor
--> see illustration

resistorpack_VSCVX.png

Absolutely right.
With this modification, the source impedance seen by the capacitor is dropped from 22.5kohms to 15kohms, so cutoff frequency will go up from 1.5 kHz to 2.25 kHz.
It will also sound louder without this resistor.

This is closer to the original, as what the patent says the cutoff frequency is about 3000 Hz with about 5nF capacitance.

But the patent numbers are inconsistent. They say the R=100k and 2R=200k, so R2R network impedance is 100k, and at the R2R network output, there's the extra 15k resistor to ground. This will both attenuate the R2R signal with a factor of 15/115=0.13, and bring down the source impedance seen by the cap to 15k||100k or about 13kohms. 13kohm and 5nF is 2.4 kHz filter, so 13kohm and 4nF would be 3.0kHz. So my guess is, the Covox cap may not be a 4.7nF but near it, someone should read the capacitor markings.

So, in light of this, it should be pretty close in theory, or even closer if the 4.7nF cap is replaced with two 2.2nF caps in parallel to get 4.4nF in total. If ceramic caps are used, my recommendation is to use NP0/C0G type if possible.

Then, the only differences left are the amplitude as dreamblaster's schematic does not attenuate the signal at all like the Covox original, so the output should be much louder in theory - maybe it is when the extra resistor is removed. Also because the R2R network resistances are lower, it presents somewhat heavier load on LPT data pins than original (mind you that the original IBM parallel port just directly drives the data bus with 74LS374 chip and it really is kind of weak to source much current into a load).

So anyway, they are different, so they can't sound identical even with same PC and same amplifier, but close enough anyway.

dreamblaster wrote:

CVX has 1 additional resistor R19, we should test by shorting this resistor

Okay, I was able to do that this evening (surface-mount resistors are smaller that I realized!) and made some recordings:

ftp://ftp.oldskool.org/pub/misc/temp/LPTDAC_c … al%20dream.flac
ftp://ftp.oldskool.org/pub/misc/temp/LPTDAC_c … t%20corpus.flac

The good news: It sounds brighter than with R19 in place.
The bad news: It still isn't near the clarity of the original Speech Thing. There's definitely some muddled frequencies, and an unintentional low-pass filter. Also, the output is louder (as expected), but it's almost too loud, louder than typical line-level output.

I could shunt the same resistor on the CVX-1 and make recordings, but since it also uses 7.5k/15k resistors, I don't think it will sound any better. Maybe it is time to change the design to use 100k resistors? I know the output level with the original design/patent is low, but it really sounds great (and looks great; bring the audio files into an editor with an FFT display and you can see the source frequencies are well-represented). Maybe it's possible to include a small amplifier on the board, if the goal is to output at line-level?

MobyGamer wrote:

Maybe it is time to change the design to use 100k resistors?

I received a version with 100k resistors. Because it was extremely soft, I haven't made a recording yet (I needed so much preamplification that there was a constant low hum, not coming from the CVX-1 itself).
But with the new information that one of the resistors should be removed/shorted, which increases volume, I can give it another shot.