On the stream yesterday Matt mentioned synthesised sound. This intrigues me a lot because in theory you would be able to recreate the sound of a locomotive with more perfection than an audio recording. In theory you should be able to fool a connaisseur of engine sounds with a synthesised engine sound. I say in theory because you will need a connoisseur of these sounds to produce them. Don't be fooled by thinking an original sound recording will always be better than a synthesised version. Would you mind that you know the sound is not real if you really can't hear the difference? Synthesised has the advantage that it doesn't need to use large audio files, or have imperfections that can come with a mic recording Tip for Matt. Start training AI to recognise train engine sounds and learn the AI how to recreate these sounds. Keep training it and will deliver better train engine sounds than a human will do eventually
In an ideal world we would have the real sounds recorded and suitably edited. However especially for retro and heritage stuff this may not be possible in which case a good synthesised mix based on a recording or YT clip as reference would be better than aliasing to the wrong sounds (looking at N3V where so many Trainz locos still emulate an Aussie Alco, even DMU's!).
It doesn't need to be an either/or. If the sound is well designed and well mixed you won't know where it came from. If you believe it's real, it really doesn't matter whether it's from 100 microphones on the real thing or a room full a chimps hitting metallic detritus rhythmically.
I know but I left the option for both out because I would suspect it would get the most votes. Just curious how people perceive artificial sound when they are in the know.
I think that synthetic sounds carry a risk that real ones don't - they could be totally off. Yeah, real sounds can be ruined with poor mixing and that - but you need the sounds to sound good and realistic to start with. The best source for recreating something is the real thing, or so I've found in the past (not developing trains, other things). However, Artificial sounds can help when a sound is unavailable (like the TGV Horn was), however I don't think I'm the largest fan of it as a mainstream move.
The main problem with artificial sound is that it misses out the mechanical element, and in trains there are lots of those. How do you artificially reproduce the sound of an engine down, then struggling to wind up again when the power is applied mid wind down and you're just starting up a hill? Would just sound a bit odd... As Protagnonist said, it's a shame you left out "a blend of both because... reasons" because that would have led to the conversation about those reasons. So for "simple sounds", digital has the edge, but some things are just darn hard to codify. Hopefully the new sound guy has experience of such things and can bring the sounds in game on from where they are (ie patchy)
Pure artificial sound will not work well. The methods that should be used, if we ever have the ability to do it in-game, would be what's the base of sample based instrument libraries is. Like modern orchestral sound libs. You have a base sample, recorded from the original in lots of variants, and then push it through a chain of synthesis algorithms. Best use for trains would be the granular synthesis of course. Sounds easy to do, but is not really, not for sounds we need for trains. Basically a simple type of that procedure is already used. You all like AP? Yes? AP does synthesise lots of sounds from original recordings using the granular methods with stopping the playback (or lowering playback speed) to get a loop out of basically not loop-able sources. Best latest example is the 387 traction motor sound. If you have a listen to the individual samples that are used, you can clearly hear out the IRCAM granular algorithm. The sounds are then layered back again what is a huge pile of work at all. The latter is why it is not done often this way. It consumes massive amounts of time. And sometimes, especially with diesel engine sound, it is not possible to do because it becomes sounding massively synthetic very fast. Of course, with lots of money, lots of time and skilled programmers there would be a chance to get a train-sound capable synthesis plugin into the game that can be used with ease then. Lets hope DTG will some day think about doing it. Or implementing the middleware approach. FMOD &co can do the stuff we need out of the box but with huge expensives on it.
I added the other option But I left it out initially just to see if there is a clear divide if you make it VS and not both. Its not a suggestion topic. I'm totally not suggesting the game should switch to synthesised audio only. It's only I see a future in audio where fake trumps real just because of the reasons you point out. Recordings are not mechanical they always sound very digital when repeated. A good example is the rattle on the Electrostar on London Commuter. I don't know if it is a recording or something synthesised but that is supposed to be a mechanical sound. When you drive that train for a long time it doesn't sound anything mechanical because the sound is always the same. A loop will always be a repeat, You can do some variations. But it is basically the same sound over and over again. With synthesised you can get it to sound mechanical because it never has to be same sound over and over again. Whatever is rattling on the Electrostar could make a different noise every time you hear it. But you will always be able to the distinguish it as the same loose part somewhere on the loco You can get more nuance in the sound of a mechanical piece when it is synthesised that would make it sound more mechanical than when you use a loop.
My take on synthesised sounds is to look at the sky box as a guide: synthesis needs a lot of skill to get good results. And train sounds have subtle sources that recording can pick up but a human may not think of when synthesising something, even if the source of the synthesised material comes from a recording When the games clouds begin to increase immersion, then I’d say that particular synthetic approach has reached maturity. It’s a long road to reach that point.
Exactly the human ear does not detect everything, AI can detect frequencies a human will not detect but that are essential to get the correct sound
Interesting topic and I can see a time where it goes to fully synthetic sounds, Would it bother me? Not in the slightest, as long as it improves an already tried and tested method that will probably have reached its limit. The thing with sound is that it changes dependent on other factors, proximity, ambience, location, direction, etc
