PhilsComputerLab wrote:

What about a variable capacitor? So one can tune it to their liking?

In theory yes.
Practically, variable/trimmer caps are usually in picofarad, generally up to 20pF, max 100pF, or maybe at most 500pF == 0.5nF.
So capacitance values are way too small, to be useful for our covox adapter.

Ok no worries 😀

If the output were buffered with a voltage follower opamp, a trim pot could be used with the cap to tweak the cut-off frequency.

I've received mine yesterday, and did a quick test on my 286 with Crystal Dream. I was positively surprised by the sound quality. It's much better than the one I built myself many years ago.
The DAC sounds very clean, almost no hiss or 'noise' coming through from the PC, and as about little distortion as one can expect from 8-bit.

I also didn't hear anything in the sense of glitching because of timing issues or transitioning between samples. So I guess no special buffering or re-syncing of the signal is required. At least not on a 286-20 or faster... Very slow PCs can still be jittery, I suppose. The playback method is basically the same as PWM on PC speaker, at least as far as timing is concerned. Then again, because there's no carrier wave, it's probably less noticeable. With PWM you immediately hear fluctuation in the carrier wave. The Covox has no carrier wave.

It sounds better than my Sound Blaster Pro 2.0, if you ask me.
Take this old capture for example: https://www.youtube.com/watch?v=Qhym3zCa7Os
You can hear a lot of hissing and other noise coming through. Also, the samples themselves aren't all that clean. It's all a tad grainy.
Whereas the Covox sounds exactly like the earlier captures that dreamblaster made, very clean.

Thanks dreamblaster!

For the sake of allowing you to compare: Crystal Dream 2 (44KHz - lowest quality available on my P120) recorded on my cheapass R2R Covox (made on breadboard) with a 4.7nF ceramic capacitor (something that's considered a bad practice) in parallel to the output.

https://drive.google.com/file/d/0B9WAZV_Pqo7g … iew?usp=sharing

I am willing to capture Crystal Dream on real covox, and also split open one of my real Speech Things and take pictures, but only if someone other than Scali requests it. No offense to Scali 😉 but I don't want to split open my vintage hardware it if it isn't necessary for the project -- but if it IS necessary (see paragraph below) then I am happy to do it.

I think it is important to define what the goal of this project is: Is it to make a great cheap modern LPT DAC, or is it to clone the Covox? The original Covox might have undesirable properties that make it less appropriate for modern hobby use. The original Covox Speech Thing was expected to be driven at lower rates (ie. 8KHz) as evidenced by the software that came with it, as well as the year in which it was invented. If I open up my speech thing, we might find that it has a low-pass filter that cuts everything off past a certain frequency...

Scali wrote:

I've received mine yesterday, and did a quick test on my 286 with Crystal Dream. I was positively surprised by the sound quality. It's much better than the one I built myself many years ago.

Scali wrote:

It sounds better than my Sound Blaster Pro 2.0, if you ask me

Thanks Scali for testing !

MobyGamer wrote:

I am willing to capture Crystal Dream on real covox, and also split open one of my real Speech Things and take pictures, but only if someone other than Scali requests it. No offense to Scali 😉 but I don't want to split open my vintage hardware it if it isn't necessary for the project -- but if it IS necessary (see paragraph below) then I am happy to do it.

If you could capture Crystal Dream on a real covox, that would be great, thanks !!
We can compare to my videos with different cap sizes, and hear if it filters more, or less, estimate what cap size corresponds.
I don't think there is need to split it open. Especially if it does not sound better than my CVX-2, this makes no sense :
My intention is not making an exact clone of the covox, but something that but sounds as good as possible, is 100% compatible,
with passive components currently available.

dreamblaster wrote:

If you could capture Crystal Dream on a real covox, that would be great, thanks !! We can compare to my videos with different ca […]
Show full quote

If you could capture Crystal Dream on a real covox, that would be great, thanks !!
We can compare to my videos with different cap sizes, and hear if it filters more, or less, estimate what cap size corresponds.
I don't think there is need to split it open. Especially if it does not sound better than my CVX-2, this makes no sense :
My intention is not making an exact clone of the covox, but something that but sounds as good as possible, is 100% compatible,
with passive components currently available.

Okay, thanks for the clarification. I won't need to open mine up.

I tried just now to make a recording, but I'm getting a high-pitched whine from its output jack whenever any of my speech things are connected to an Athlon I had out for testing. I'm going to dig out a 486 and try with that to see if the whine is still present.

EDIT: I tried it in a 486 and the whine is still there once it is connected, but it is much less noticeable, likely due to better shielding. I will put together a quick reference video of Crystal Dream and reply back with the URL.

I also got a package today 😊

... and after some soldering I have these now :

These are 2 CVX2 LPTDACs, one with resistors that have 0.1% tolerance - the other with 1% tolerance all picked by dreamblaster from the same batch (so they can be as similar to each other as possible). I have installed a temporary capacitor "socket" so it is easy to test different values. It's not good for a permanent solution though...

So naturally I made some Crystal Dreams recordings:

0.1% CVX2 with 4.7nf capacitor: https://soundcloud.com/keropi666/serdaco-cvx2 … -crystal-dreams
0.1% CVX2 with 10nf capacitor: https://soundcloud.com/keropi666/serdaco-cvx2 … -crystal-dreams

1% CVX2 with 4.7nf capacitor: https://soundcloud.com/keropi666/serdaco-cvx2 … -crystal-dreams
1% CVX2 with 10nf capacitor: https://soundcloud.com/keropi666/serdaco-cvx2 … -crystal-dreams

I cannot hear a difference between the 0.1% and 1% versions with the same value capacitor - since even 1% tolerance is really good and the resistors are from the same batch I don't think it really matters what resistor tolerance one uses.

CVX2 blows away my previous covox clone - infact I'll go as far as to say that the previous clone is almost garbage compared to CVX2 🤣

Thanks so much dreamblaster for creating and sending over the CVX2! It really makes you spin your head with the sound quality it produces over the LPT port 😎

ps. anyone knows a program that allows 2x LPTDACs to be used for stereo sound?

keropi wrote:

ps. anyone knows a program that allows 2x LPTDACs to be used for stereo sound?

FastTracker II has a "SoundPlayer LPT1+LPT2" option, which should work.
Not sure if CD and CD2 also support this. THey do have a stereo SoundPlayer option, but you can only choose one port, I think... Might be the for latched version on a single port.

nice thanks for the FTII tip Scali. Now to see how to install a 2nd LPT on my p1 build for science reasons 🤣

Okay, I had a chance to make a video covering the versions of the Covox Speech Thing that I have, and I recorded Crystal Dream output from it:

https://www.youtube.com/watch?v=M0kxcc67l0w

Here's the raw audio file without any processing: ftp://ftp.oldskool.org/pub/drivers/Covox/Spee … ch%20Thing.flac

Spoiler alert: A real Speech Thing sounds extremely good. In fact, I personally think it sounds the best out of all the samples provided thus far in this thread. So, how to clone its quality? Why not just clone it directly, since the patent has expired: https://www.google.com/patents/US4812847

Things I learned from the patent:

• It is indeed a resistor ladder (called a "resistor array" in the patent). The exact values are in the patent, and the diagrams are helpful.
• The patent describes a capacitor value that results in a low-pass filter of 3 KHz, except that in my recording you can clearly hear frequencies above that so it's possible they removed that from later iterations of the hardware. It's also possible that part of the invention is inside the speaker that comes with a Speech Thing.
• The original application was meant to be an output sample rate of 8KHz.

Great video MobyGamer, Thanks!
Indeed the original sounds the best, but the CVX-2 sounds just as good or at least much better than other clones.

Nice video indeed! and there is a difference in sound too - but just to make sure you are recording with your soundcard's mixer/processing disabled right?
I think you'll have to open one of these speech things 😁

MobyGamer wrote:

Spoiler alert: A real Speech Thing sounds extremely good.

Thanks for the video!
And indeed, it sounds excellent. Most clones don't do it justice at all.
I think the most apparent is at the start of the demo, when the synth strings come in. They sound a bit noisy on the CVX-2, but very clean on the real Covox.
Again when the second part of the demo starts, the synths are really clean on the real Covox, and somewhat noisy on the CVX-2.
I wonder what is causing that. It seems that the 100nF cap filtered some of that noise out... but on the whole it sounds more muted than the Covox does... so hmmmm.

MobyGamer wrote:

It's also possible that part of the invention is inside the speaker that comes with a Speech Thing.

It's also possible that they included more circuitry than just the part that is in the patent. Or that they modified it somewhat over the years.
Did you try all three, to see if there is any difference in how they sound?

Also, I see what you mean, the case doesn't seem to have any screws or anything. Is it glued shut or something?

^ Nope not glued thanfully! There are clips on the side so it's just a matter of lifting them so the shell separates . But there is no guarantee the insides are not covered with epoxy/glue - I hope they are exposed.

I want to ad that the original Covox sounds like it has no filtering cap at all when compared to the CVX2 with no cap, they sound identical.
But the CVX2 sounds like it has lower bitdepth which is directly responsible for "dirt" behind the sound or self noise level.

Listen carefully for the difference in noise at the break part:
https://youtu.be/bn-spjT_A-A?t=49s
https://youtu.be/M0kxcc67l0w?t=6m32s

Scali wrote:

Thanks for the video! And indeed, it sounds excellent. Most clones don't do it justice at all. I think the most apparent is at t […]
Show full quote

MobyGamer wrote:

Spoiler alert: A real Speech Thing sounds extremely good.

Thanks for the video!
And indeed, it sounds excellent. Most clones don't do it justice at all.
I think the most apparent is at the start of the demo, when the synth strings come in. They sound a bit noisy on the CVX-2, but very clean on the real Covox.

It's still not as clean as an Amiga. That's the quality you should be aiming for IMO. Preceding the R2R ladder with a latch powered by a low noise supply might help improve things.

are you seriously comparing an LPTDAC device with Paula? c'mon man, this is not a dedicated DAC we are talking about, this is as simple as it gets - don't expect it to replace your soundblaster 🤣 🤣 🤣

lissajous wrote:

It's still not as clean as an Amiga. That's the quality you should be aiming for IMO. Preceding the R2R ladder with a latch powered by a low noise supply might help improve things.

You can't compare the software mixer in Crystal Dream with an Amiga. It's an early MOD replay routine, only uses 8-bit (perhaps rounding down to less even), and doesn't use any kind of interpolation for resampling.
It's never going to sound as good as the 4 dedicated DACs in an Amiga.
If you play a single 8-bit sample on an Covox, then it could/should be comparable with a single Amiga channel playing an 8-bit sample. But my impression is that it is.