Audio quality in DOSBox with SCI Companion recordings

[troflip]

I've been noticing some very poor audio quality with the voice recordings I've been making. This only happens in DOSBox (not Scumm), but it doesn't happen with Sierra games.

It happens mainly with 'S' sounds, the sound is very crackly. It doesn't seem to be a frequency issue, because I still hear it even if I lower the frequency by several octaves. And the waveform also isn't clipping, and is roughly the same level as sounds in Sierra's audio resources. So I'm not really sure what's causing it.

Any audio experts out there that can hazard a guess?

[MusicallyInspired]

Frequency would have been my first guess. You said you tried several octaves lower. What exact settings are those? The lower the frequency the less sibilant the "s" sounds will be, obviously. I'm just wondering if it's a nonstandard setting like something between 8000, 11025, and 22050 that might be throwing off the driver. I don't know that it would throw it off, but that would be my second guess. I assume you've tried those standard settings. I haven't tested it yet myself so I don't know what it sounds like, but I'll give it a go tonight after work and see if anything is any more illuminating.

[Collector]

What driver are you using?

[troflip]

I'm using audblast.drv. I tried audthund.drv also.

Although, I'm noticing it less this morning, lol. I'm not sure I ever compared the Sierra resources in game. They do have some light crackliness too. I tried recording the same speech at roughly the same levels as a Sierra speech resource, and the difference is not that great.

When I said I lowered the frequency several octaves, I just mean I did a non-duration-preserving pitch shift in my audio editing app. So it just stretched things out. A couple of octaves should have brought the highest frequency down from 11KHz to about 2.75Hkz (for a 22Khz sample).

The amplitude waveforms (mine vs Sierra's) "look" quite different. Sierra's appear much "fuller", if that makes sense. I also tried looking at them in a spectrum analyzer and they looked quite different.

I'll do some more experiments.

[troflip]

Here's a recording I did of "The plants say nothing" from KQ6. I've attached the original, and my recording. I used some 10:1 compression to even out my levels, and ran a de-esser to reduce sibilants. They are both about the same level of loudness.

But my recording causes the Sierra interpreter to crackle in DOSBox.

Also attached is a spectrum analysis. Horizontal axis is time, Vertical is frequency. The S's are clearly noticeable, those are the orange bits at the top (the Sierra recording has a pause between "plants" and "say", whereas I just slurred them together. The frequencies are distributed a little differently, but that's probably just because of the difference in our voices?

[MusicallyInspired]

No doubt. The gaps are probably also due to a gate effect before the compression, on the Sierra recordings.