That is correct.

Here’s why… the parameters that make up the Voice (less the oscillator) each take up about 2KB of data.
The data that make up the oscillator (that is the audio samples and Waveforms) can be huge by comparison. The Chick Corea Mark V is over 400MB of audio data. The Bosendorfer Imperial Grand is over 400MB of audio data.

Each of these Libraries has Voices that point to the audio data. The Voice data for each Voice is about 2,000 bytes of data. The Oscillator that they all use, is 400,000,000 bytes of data!

The rest of the VOICE parameters (those that involve Filters, EGs, LFOs, Effects, etc., etc.) are loaded to the User Voice locations.
Within the Voice, each Element has an OSC > WAVE screen that “points” to an item on the Waveform List.

There are two Waveform Lists: one for Presets, the other for User (Flash Board).

There are 3,977 Preset Waveforms and room for another 2,048 User Waveforms that can be stored on the Flash Board.
Not all Libraries are like the Chick Corea Mark V, and the Bosendorfer — very detailed recreations of one instrument.
The EDM Library data, for example, is just over 53MB and is made up of multiple instruments and sound effects, of varying sizes

The audio data is humongous compared to the Voice parameters that shape it into a playable synth sound.
This is why the whole separate Flash memory thing exists. While it takes about a minute to load 384 Voices, 256 Performances, a bunch of Songs and Patterns… it can take more than twenty minutes to load a Flash Board full of Waveforms.

Thousand versus Millions is no joke.
You’ve held $2,000.00 no doubt, but you probably have never held $53,000,000.00
Counting to 2,000 is doable, counting to a million, is something you laugh at…
So loading a bunch a sounds that are small data size is doable… waiting for millions of bytes of data to load is something you want to avoid.

But once installed on the Flash board you never have to wait for them to load again. They (the Waveforms and Samples) remain in place, exactly like the Preset Waveforms and Samples installed by Yamaha at the factory. You never have to wait for them to load. When you load a Bank of User Voices that reference the Preset Waveforms, you do not have wait for audio data to load… the same will now be true of Voices that point to Flash Board Waveforms.

When you want to load your own files that use custom Waveforms, you load “without Waveform” - the Voices, Performances, Songs and Patterns load in less than a minute! Because all the BIG DATA remains in place on the Flash Board or in the Preset area.

Ok think I get it now....the voice is just the recipe sheet and the waveforms are the ingredients.