Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Matt]

We added a DSP called 'Bitdepth simulator' to MC18.

It's a way to artificially reduce the bitdepth of the signal inside Media Center.  It's useful for understanding bitdepth, for testing, or just as a toy.

So for example, you could test and see if you can really hear a difference if you play a 24bit file back at 16bit instead of 24bit.

Or you can figure out what's the lowest bitdepth at which you can tell a difference.  Pick a number, say 10bit and turn the DSP on and off and see if you can hear a difference.  Then add a bit and see if you can still hear a difference.  

For me, 12bit is audibly degraded.  13bit requires concentration, and I can only hear a difference sometimes (using headphones and a high volume).  After that I'm not sure I can pick above 50% (eyes closed, tapping the space bar in DSP Studio to toggle).

To play with this yourself, make sure you have MC18.  Then look in DSP Studio > Parametric Equalizer > Add... > Bitdepth Simulator.  Make sure the filter is checked and that 'Parametric Equalizer' is checked in DSP Studio.

This sort of stuff is fun for an audio nerd like me.  Hopefully a few of you will find it interesting as well.

[InflatableMouse]

Cool.

What would make this even cooler is if you could build an ABX tester around it. Let MC control the bitrates and record YES/NO choices without the listener knowing what it is he's actually listening to, only let him switch to the original to compare. Let's say you start it at 10 bits and have it gradually increase the bitrate once confidence passes 95%.

The thing is, when you know what you're listening to, you will imagine a difference whether you want it or not. It could very well be that even that 13 bits is transparent for you if you don't know that you're listening to it.

I did an extensive ABX test with mp3's one time with several people. We could not distinguish Lame 192kbps VBR and upward from the original. If it weren't for 1 or 2 songs for which the soundstage was screwed up, 160kbps VBR would have been sufficient. I still rip my CD's to Flac though and I'm a fan of 24/192 . It's a lot of fun to take these tests but doing it manually is just too much work.

[Matt]

The thing is, when you know what you're listening to, you will imagine a difference whether you want it or not. It could very well be that even that 13 bits is transparent for you if you don't know that you're listening to it.

That's totally true.

A poor man's test is to close your eyes, and mash a key a bunch of times (next track for MP3/Lossless or in this case the spacebar for the bitdepth toggle).  You won't know what's playing.  Then listen with your eyes closed and toggle until you pick a favorite.  Open your eyes and see if you're right.  Repeat a few times.

If you're honest with yourself, you'll have to say "boy, I can't really tell."  Or if you think you really can tell and open your eyes and are wrong, it sort of proves you can't.

But now for the philosophy question: if knowing something makes it sound better, does that mean illusion matters?

[InflatableMouse]

But now for the philosophy question: if knowing something makes it sound better, does that mean illusion matters?

I think we can better answer it if we turn it around: if knowing something makes it sound worse, does it matter? Then the answer is definitely yes.

For me it's about peace of mind. If there is no reason for me to suspect bitrate, then theres no reason for me to question it.

[audioriver]

Any plans to implement bitdepth custom selection the way you did with sample rate? "Source bitdepth" causes problems on lossy files (goes to unsupported -for many cards- 32-bit playback mode).

[Vincent Kars]

I assume the output is dithered, correct?

[audioriver]

I assume the output is dithered, correct?

Not sure what you mean. I've set the bitdepth output to 24-bit, source bitdepth threw playback errors. However, I've tested some MP3's and they now play fine with source bitdepth (it's set automatically to 24bit)... Guess I haven't tested after upgrading to MC18 and missed some changes. (or the audio driver behaves better in Win 8?)

[Vincent Kars]

Tried it with a 24 bit recording.
Set bit depth to 16, could not hear any difference.
At 14 bits there is a slight noise in the background but I don’t think the difference will survive a unsighted test in my case.
Nice toy!

[AndrewFG]

But now for the philosophy question: if knowing something makes it sound better, does that mean illusion matters?

Aaah. Is it philosophy ... or is it reality? ...

The listening process is a kind of DSP process that runs on "squidgy-ware" in the brain. The process involves an ADC conversion from the analog vibrations of little hairs in the ear into digital neuron firings. It then does something like a Fast Fourier Transfrom to convert the firings into the frequency domain. The frequencies are split into silos, and the silos are parallel processed, analysed and compared with stored references in memory and thus mapped into percieved voices or instruments. The voices and instruments are also processed binaurally, and mapped into a stereo soundfile image. (Etc. Etc.) This is a seriously complex DSP with many inputs, parameters, knobs, and whistles; some of which are driven by physics and some of which are driven by chemistry (emotions). So if something makes it sound better, who cares if that something is down to physics or chemistry...   ;-)

[sla]

The fact that one cannot hear difference does not mean that there is no difference to hear.

Once should know the material very well and that material should vary a lot [check a well recorded piano music], and of course it also the matter of the equipment [pre-amp,amp,speakers,etc...].
Secondly, for instance in my opinion DSD is slightly better then PCM but I could live with PCM... so it depends what kind of difference you are looking for.

My 2 cents..

[J-a-k-e]

So for each 'bit' you reduce the bit depth by from the original bit-depth you loose 6 db of dynamic range. So if you reduce the bit depth of a song that uses the full dynamic range, the quietest parts will fall into that static fuzz noise you get. This is most noticeable with well recorded music that has a very quiet lead in when you reduce the bit depth and turn it right up.

[Trumpetguy]

Aaah. Is it philosophy ... or is it reality? ...

The listening process is a kind of DSP process that runs on "squidgy-ware" in the brain. The process involves an ADC conversion from the analog vibrations of little hairs in the ear into digital neuron firings. It then does something like a Fast Fourier Transfrom to convert the firings into the frequency domain. The frequencies are split into silos, and the silos are parallel processed, analysed and compared with stored references in memory and thus mapped into percieved voices or instruments. The voices and instruments are also processed binaurally, and mapped into a stereo soundfile image. (Etc. Etc.) This is a seriously complex DSP with many inputs, parameters, knobs, and whistles; some of which are driven by physics and some of which are driven by chemistry (emotions). So if something makes it sound better, who cares if that something is down to physics or chemistry...   ;-)

Slightly off topic, but too fun not to be discussed. It may very well be that the brain works in the frequency domain, but the time domain is very, very important for perception as well. The characterization term "timbre" can not easily be explained without including time. One simple example - playing the same tone with an accenture with an oboe and a trumpet will create two similar, but still easily distinguishable sounds. If you remove the first few milliseconds of each of the two sounds (the accented attack), and only play the steady state part of the tones, they sound almost identical. This attack part can theoretically be transformed into the frequency domain, but I am not sure if any power spectrum is useful for describing how we perceive the sound.