Skipback SCAM-pling!

I will start with the warning that I am a newbie on my Fantom as well, so others will probably have much better solutions to this, but I too noticed that problem (if it even really is a "problem") and I was able to work around it pretty easily by going into Sample Edit and selecting the Amp function which allowed me to dial up the volume on a couple of samples to where I needed them to be. I'd love to hear if there are any better ways of doing this.

You can "Normalize" the sample. This will increase it's volume to the max possible. You can then turn the samples volume setting down to get what you need.

i might be absolutely the wrong guy because I dont own a fantom. but isn't it possible that not your fantoms but your monitor-systems are the reason for this volume-drop?

if you play and hear your fantom, you route usually the signal i.e. to a mixer/monitoring system. but when your fantom resample something, this amplification does not exist and thats why you hear the resampled sound less loud???

you could figure this out by using headphones instead of your monitors.....

good luck

urs

This is probably the single most common observation/complaint of new Roland XV-based engine sampler users.

I have a working theory as to why this happens. I believe the volume difference stems from a difference in bit depth between the internal sound processing system, which is 24-bit, and the sample recording system, which is 16-bit. It seems that the way the Fantom sampler handles this is by simply dropping the 8 least significant bits of each 24-bit data byte from the internal sound generator (just from my observations, I don't know for sure this is what's happening). This has the effect of significantly reducing the volume of resampled audio, but also prevents clipping for exceptionally loud sounds. I believe Roland chose to use this method as opposed to resource-expensive dithering to save CPU power for sound generation and other tasks.

In response to user concerns regarding this issue, Roland has included some parameters to help you compensate for this volume difference in the latest Fantom-X OS (2.10). You'll notice that in the Sample Parameters screen you have a sample gain. You can increase this to help compensate for low sample volume without having to normalize the sample. This parameter is saved along with the sample when you save it to USER or CARD storage area. Also, if you're assigning the sample to a Tone in a Patch, you can increase the Wave Gain for that Tone.

Regards,

Nathan Sheldon
http://www.nathansheldon.com/

Nathan may well be right but it would be nice to have an absolute description of this. I'll make a note to get in touch with Roland UK.

I certainly hope it's not truncating sample bit length, this would impact sound quality considerably and I don't understand why that would effect volume levels. Truncating the sample gives an entirely different result to dithering (effectilvely resampling at 16 bit). I think it's more to do with the DSP dithering and perhaps filtering which cause the problem. As Nathan said, it may just be there inherently to avoid digital clipping - which is nasty.

The problem with normalising is that is completely loses all dynamics - not that this is a problem for most music that comes out these days, louder=better - yeah, my arse, pass me the aspirin.

Steve

Hi Steve.

"I certainly hope it's not truncating sample bit length, this would impact sound quality considerably and I don't understand why that would effect volume levels."

Truncating bits would affect sound quality in the following ways:

* Dynamic range would be reduced to a 16-bit level (96 dB max).
* S/N may be reduced (to somewhere in the 90 dB range)

I wouldn't personally call this a considerable impact on sound quality, but some may. These values are lower than a 24-bit audio signal which have a theoretical dynamic range limit of 144.5 dB and a S/N of about 120 dB.

The reason these are the only things that are affected by dropping the 8 least significant bits is because by doing this the volume is simply being reduced so that the loudest sound can only be 96 dB. But again, I'm not certain this is how Roland is doing things in this particular situation, but if I were a hardware engineer with limited processing resources (the Fantom CPU is only a 120 MHz Hitachi SSH2 CPU, I've seen the service manual that shows this to be true) this is probably how I'd do it rather than wasting cycles applying a complex dithering algorithm that would reduce polyphony or sequencer accuracy.

Regards,

Nathan Sheldon
http://www.nathansheldon.com/

Hi Nathan!

Just thought I'd be a smart-alec and mention that the CPU is actually an SH-3/7706 running at 128MHz. You can kick me now. :>)

http://www.piaknow.com

Haha, right you are piaknowguy. A second look at the Service Notes confirms this.

Steve, I think I forgot to answer your question as to how dropping the 8 least significant bits would affect volume. Let me try to explain it briefly.

Let's say you have a hot 24-bit signal coming from the Fantom sound generator. The highest audio sound waveform peak gets to a sampled level of 16,777,215 which is

1111 1111 1111 1111 1111 1111

in binary. Those 24 bits aren't going to fit into a 16 bit word so we drop the last 8 bits to get

1111 1111 1111 1111

(which is the same as 65,535 in decimal).

Now the bottom of the valley in the audio waveform coming from the Fantom is, say, 1000 in decimal (accounting for a little noise), or

0000 0000 0000 0011 1110 1000

in binary. This truncates to a 16 bit number of

0000 0000 0000 0011

or 3 in decimal. So now you go to play back this 16 bit sample that you just created from resampling the Fantom. So what happens to the 16 bit signal when it's sent back through that 24 bit processor and 24 bit DAC? The 16 bit high of

1111 1111 1111 1111 is read as 0000 0000 1111 1111 1111 1111

by the 24 bit processor, and the 16 bit low value of

0000 0000 0000 0011 is read as 0000 0000 0000 0000 0000 0011

by the 24 bit processor. These are obviously lower values than the original signal. So why not just putt 1111 1111 or 0000 0000 at the end of the 16 bit data to increase the volume? For a few reasons. 1) the internal samples in the Fantom are 16 bit as well. What happens when you play 128 very loud 16 bit sounds simultaneously? If your processor is a 16 bit processor, you get sever clipping of the output audio. If you were to pad the 16 bit sample with 1111 1111 or 0000 0000 at the end of the 16 bit word, you'd get a louder sound, but you'd also risk over-driving the 24 bit audio processor and 24 bit DACs, causing clipping.

I don't know if I explained that well, and I'm still not positive that Roland actually does this exactly as I've surmised here, but it does seem to fit my personal observations anyway.

Regards,

Nathan Sheldon
http://www.nathansheldon.com/

Hi Nathan!

Just thought I'd be a smart-alec and mention that the CPU is actually an SH-3/7706 running at 128MHz. You can kick me now. :>)

Damn, I thought the cpu would be faster than the v-synths cpu since it is dealing with multiple ttracks... ours is running 330 somethin mhz... correct me if I'm wrong.

http://www.myspace.com/11505663

Hi anti_pulse.

The V-Synth uses a variant of the Hitachi SH-4 32-bit CPU at 266 MHz and also utilizes an Toshiba variation of the PowerPC chip (TC200 series) running at 133 MHz for the various data interfaces and is slaved to what appears to be a custom Toshiba TC223 DSP (also running at 133 MHz).

In all, the V-Synth has significantly more processing power than the Fantom-X. The reason it needs it is basically because of the sophisticated real-time sound manipulation functions. Manipulating pitch without affecting tempo, or vice versa, in real-time requires significant data processing (even with pre-analized waveforms). Users of the V-Synth and the Fantom-X have probably noted the significantly higher quality of pitch and/or time manipulation of sounds on the V-Synth over the Fantom-X. This level of quality requires greater processing power. Adding sophisticated physical modeling algorithms (COSM) also adds significant processing requirements.

In any case, none of that's really all that important as long as you can make inspiring music with it.

Regards,

Nathan Sheldon
http://www.nathansheldon.com/

If they're doing bit truncation scullduggery, apart from a good kicking (in a loving way), I think there could be some merit in recording the outputs (digitally) assuming they "bother" to dither the ouputs of course!

My Fantom is coming in the next few days. I thought resample/skipback was the obvious choice, but I can see I need to do some serious empirical testing using the actual outputs.

Sorry but this is the sort of technical stuff that annoys me - when a manufacturer disguises the issue by pretending everything in the digital domain must be perfect. Lossless - my 3 lettered donkey!

AFAIC normalising increasees the noise floor by an an equal amount and obviously does nohing to address resolution. There could be similar issues with digital gain using the amp section.

Sorry but the level shouldn't drop and they shouldn't be bodging it. And I say that with my software programming hat on.

JMTCW.

Hi midiwhale.

If you haven't already been welcomed, then welcome to the Roland Clan.

Yes, normalizing (and using the amp function) will increase the noise floor in the same amount as it increases the signal. Yes, this is unfortunate and a good dithering mechanism for resampling (or any sampling, for that matter, from a 24-bit source to a 16-bit destination) would be far preferable to to effectively dropping the volume. It's either that or sample everything at 24-bit.

However, I understand why Roland did this. You cannot mix multiple 24-bit signals and guarantee that there won't be any clipping. However, you CAN mix 128 separate 16-bit sources (128 voice polyphony), add effects, and guarantee that even if all 128 sources peaked at maximum volume all at once, they wouldn't overdrive a 24-bit data path. This is why they didn't make the sampler 24-bit (if 128 simultaneous 24-bit signals peaked at once, you'd get massive clipping). And the reason they don't use dithering is because the 128 MHz SH-3 CPU and custom WX DSP just aren't powerful enough to do dithering and all the other Fantom functions at the same time. You'd either loose polyphony, loose sequencer resolution, loose screen updates altogether, or a combination of all of those, none of which are acceptable just to resample with dithering.

Now, don't get me wrong. I personally think that Roland could have used a faster CPU instead of the SH-3 (perhaps the SH-4 like they're using in the V-Synth), considering how much they're charging for the Fantom-X. This would have likely allowed them to do this dithering in real-time. But Roland's not just an instrument manufacturer, they're a business, and sometimes design sacrifices are needed to get an acceptable profit from a product, at a price point that people will pay.

Oh, and the Fantom-X digital output is 24-bit, 44.1 kHz fixed, strait from the WX DSP sound source. The digital input accepts all S/PDIF standard bit depths of 16-, 20-, and 24-bit (all only at 44.1 kHz though), but all are converted to 16-bit depth when sampled (I haven't tested digital input levels to see if levels from a 20- or 24-bit source are quieter than expected).

Regards,

Nathan Sheldon
http://www.nathansheldon.com/

Nate; thanks for the descriptions. I see what you're trying to say but I'm unconvinced at the moment! You may well be right, but I have a problem with truncation, since a 24 bit waveform truncated to 16 bit won't sound the same as resampled (to 16 bit) 24 bit waveform - which is what I think is going on. The first 16 bits do not describe the sound of a 24 bit waveform, hence I'm sure you would hear significant tonal changes - not just volume.

I'm going to put a mail across to Roland UK and see what they say. Plus they can update me on the sample converter, have you got invloved in this in anyway?

Steve

But isn't it about resolution? Sure, 16 bits cannot fully describe a 24 bit waveform, but can offer adequate resolution for most ears.