Song sounds perfect in ableton but exported it sounds bad ):

[orin]

Thanks for taking a look, pottering!

Now, please allow me to raise the bar on what to expect from a DAW. When you export an image from Photoshop (or any imaging program), and then export the same again, you expect the difference to be zero: black. Likewise, when you export sound from Live (or any music program), and then export the same again, you *should* expect the difference to be zero: silence. Every sample should be bytewise identical.

Why? Because you can have amazingly random-seeming analog emulations, "free" running LFOs, and all the lot of complex "unsynced" effects, without one bit of true randomness. You can get sample-by-sample identical exports with all such effects perfectly expressed in a 100% deterministic way because randomness in software is actually "pseudo-random", produced by transforming original state (the "seed") from one random number generation to the next. All you have to do is properly manage the RNG state in the host. Similarly, the "free running" processors like non-tempo-synced LFOs simply need to be initialized consistently to produce the same signal. This is an extremely valuable property of both software and art created with software.

If you don't manage state in this deterministic manner, your output is effectively randomized! What if your image exports varied depending on the exact clock-time you saved the PNG? Artists would be up in arms about it. But for some reason, musicians seem to think it's to be expected, or even desirable, that sound renderings vary with chance effects like time and LFO phase.

I'm trying to help Live gain control over audio rendering, for the sake of consistency and reliable music-making. But if musicians just expect and accept unreliable renderings, then what are the odds?

By the way, jestermgee, science doesn't require video recordings or downloads -- repeatability is the essential quality. My instructions are clear and the experiment is simple. It even works with the stock "Ensemble Brass" instrument: it takes 2 minutes tops. Try it yourself to be convinced that this is not user error. It may well be Analog, as pottering hints, but in any case it is pure Ableton software, and it is inconsistent.

[Fanu]

Hi, I am using Ableton 10 Lite
…

Try previewing the file using the preview feature you get using space bar – does it now sound different?

[Da hand]

Hi, I am using Ableton 10 Lite, ....

Hi Stennos, a few things to consider in this order:

1. What sound output (soundcard driver and physical output(s)) is used in Live settings vs iTunes settings? Is it the same?

2a. In Live, what are your Master output levels in Live before exporting? What have you set the Master Fader levels to? and what is the actual dBFS readout on Live's Master Fader meter?

2b. Are your Master levels going in the red (above 0 dBFS) before exporting?
2c. Are you using a limiter on your Master track?

3a. Do you have any effects on in iTunes? Audio or otherwise? for example an EQ boosting bass or some sample conversion
3b. If you preview the track in another software other than iTunes, are you still having the same problem?

[jestermgee]

By the way, jestermgee, science doesn't require video recordings or downloads -- repeatability is the essential quality. My instructions are clear and the experiment is simple. It even works with the stock "Ensemble Brass" instrument: it takes 2 minutes tops. Try it yourself to be convinced that this is not user error. It may well be Analog, as pottering hints, but in any case it is pure Ableton software, and it is inconsistent.

Hey, if you think the best way to prove your theory is to ask others to setup their own experiments rather than provide your own findings then good luck. I'm not about to take a new poster serious in their claims that the software doesn't work when they cannot offer even a simple test file for download to save someone else the time to setup the same experiment (who was willing to download the file and actually test and confirm mind you). I couldn't be arsed to setup experiments when I know there is no issue through my own 10+ years of using Live.

As has been pointed out, your test was using something that has free running oscillators and LFOs designed to offer a bit of randomness in the output. And your complaint is that it is behaving in a random way?

I get that you feel that even though it is "random" it should still produce the exact same result but there are ways to program a patch so that DOESN'T happen for the same reason people like to use Analog gear or real musicians... it gives you a somewhat random output each time which may actually be desirable and you would probably find that with some knowledge you could actually tame that if you needed too using key tracking or something so it always resets the LFOs on a keypress rather than being more free running so that it does produce the same result from start. But that requires some knowledge on how things are working. You will probably find there are effects for Photoshop or After Effects that can also produce a random effect each time for the same reasons. Maybe you need 10 of the same render but all slightly different?

I'm not sure your argument is going to win too many over there. Again, sounds like a case of user error and again, try your test with a simple waveform experiment and null the results. If it does not null them come back, post the Live project and someone may be able to look further into it.

I'm pretty sure this myth is busted but as the memes say, prove me wrong.

[[jur]]

Thanks for taking a look, pottering!

Now, please allow me to raise the bar on what to expect from a DAW. When you export an image from Photoshop (or any imaging program), and then export the same again, you expect the difference to be zero: black. Likewise, when you export sound from Live (or any music program), and then export the same again, you *should* expect the difference to be zero: silence. Every sample should be bytewise identical.

Why? Because you can have amazingly random-seeming analog emulations, "free" running LFOs, and all the lot of complex "unsynced" effects, without one bit of true randomness. You can get sample-by-sample identical exports with all such effects perfectly expressed in a 100% deterministic way because randomness in software is actually "pseudo-random", produced by transforming original state (the "seed") from one random number generation to the next. All you have to do is properly manage the RNG state in the host. Similarly, the "free running" processors like non-tempo-synced LFOs simply need to be initialized consistently to produce the same signal. This is an extremely valuable property of both software and art created with software.

If you don't manage state in this deterministic manner, your output is effectively randomized! What if your image exports varied depending on the exact clock-time you saved the PNG? Artists would be up in arms about it. But for some reason, musicians seem to think it's to be expected, or even desirable, that sound renderings vary with chance effects like time and LFO phase.

I'm trying to help Live gain control over audio rendering, for the sake of consistency and reliable music-making. But if musicians just expect and accept unreliable renderings, then what are the odds?

By the way, jestermgee, science doesn't require video recordings or downloads -- repeatability is the essential quality. My instructions are clear and the experiment is simple. It even works with the stock "Ensemble Brass" instrument: it takes 2 minutes tops. Try it yourself to be convinced that this is not user error. It may well be Analog, as pottering hints, but in any case it is pure Ableton software, and it is inconsistent.

Orin,
If you don't want this to happen simply use sync'ed LFO and oscillators with phase retriggering, there's no other way and it's perfectly logical.
Also avoid "analog-style" plugins that introduce random noise and saturation to feel "vintage". You just can't expect them to sound exactly the same at any given time... they're specifically designed to avoid this.
The analogy with an image just doesn't work: an image is, by definition, static. Sound is Time, there's no such thing as a frozen sound, it can't be something else that time passing, air moving.

[Stennos]

You’re hopping between two softwares without having an additional reference in either.

It’d probably help get a better idea as to whether or not things actually are being affected by doing a few of the following.

1. When you render a file, try opening it back in Live. If it sounds drastically different inside the program that rendered it, then you have a problem. If it sounds the same, then the problem is between the different programs.

2. Also in Live, try opening a good quality example of a recording done by a professional, especially if you can find something along the lines of your direction for your recordings voice-wise. You might be trying to make it sound perfect in Live only, and in the narrow circumstances of your own setup, but clearly you’re expecting others to listen to it on things like iTunes etc, and you probably want it to sound generally palatable. Comparing it to a professional recording, you can see if certain frequencies sound overbearing.

Thank you Y2D4. That is a very good idea, of course! I will get onto this straight away, and post back my findings. I understand what you mean about the overbearing frequencies, I have been listening very carefully to professional voice over recordings, BUT, in iTunes and Youtube. I have not thought of bringing them into Ableton to see what they sound like there.

[orin]

jestermgee, posting videos or files saves no time. It literally takes under 2 minutes to prove to yourself that my experiment is valid. Try it with "Ensemble Brass", a stock sound. No myths, no user errors, just two back-to-back renders that differ. Some people think it is right and well that they should differ. I disagree.

Thanks for the response, jur. Retrigger is a way to enforce phase reset per note, but there are creative uses for free LFOs and I still think these should have some fixed anchor in time -- at the very least, they should trigger (initialize the phase) at render start time, even when retrigger is disabled. Right now, their phase is uninitialized and effectively up to chance, so making use of the LFO means losing control over the sound. Tempo-synced rates are really just another way to specify the frequency: there is an equivalent value in Hz, it's just calculated.

Sure, synthesizers *can* be written to produce unpredictable outputs, but they can also be written to produce perfectly identical sample streams with the right initialization and control. The sound characteristics are not affected -- i.e., the analog simulation, the warmth and noise, the "time passing, air moving" effects can all be expressed... yet they can be expressed *exactly* the same again, if the software is engineered for control and consistency. With this design, happy accidents can be kept instead of lost. If random number generator seeds are exposed for control, you can even get different results when desired. This is roughly analogous to a phase for something that makes use of randomness. Yes, I want even a white noise generator to produce exactly the same noise stream every time, unless it is explicitly changed.

But hey, I understand this is a significant feature request, and Ableton is already hard at work on Live 10.1, which looks to be awesome, so I'll drop my case for now and just hope some seeds of ideas will grow in the minds of engineers that happen across this thread. It really is wonderful when you get a big complex system working deterministically. Every snowflake falls in its right place.

Cheers, all!

[Stennos]

FAO: Da hand, Fanu, and yur2die4, thank you.

I combined your advice and have resolved the issue as follows: I have been a bit lame in not re-calibrating since getting new equipment and software.

1. I realised from your questions that I was not working from a base reference - neither in the physical nor the digital aspects - so I decided to calibrate across the board using the iMac core audio output with all the volumes turned up to maximum. Using the Ableton tone generator, I made sure that the output to my monitor speakers was at -6db and using an external decibel meter scrolled through th frequency range and set my speaker volume to a peak of 76db (which I read is recommended). I also made sure all the settings in the "Audio MIDI setup" app were set to 24bit 44100kHz across the board.
2. I kept the Ableton input to the Microphone via the external pre-amp, but I changed the input values to the same as the export values. i.e. all the advice says that there really is no need to be working in anything other than 44100kHz and 24bit. (Latency is not an issue for me).
3. I then got a downloaded BBC audio file from Radio 4 (Melvyn Bragg and others talking) and imported this into Ableton and exported it, and then remastered it, normalised it, etc. and was nicely surprised to find that it sounds identical in Ableton, and in iTunes.
4. What I think was happening was the tracks I had been producing, were actually too quiet, and when I was playing them back in iTunes I was compensating for the quietness by raising the volume in iTunes, and that was identifying a problem in my mastering because the mix wasn’t different between iTunes and Ableton, the volume difference between the two pieces of software was enough to radically change the overall feel of the audio, as the lower frequencies became more prominent, so in fact iTunes ironically was telling me the truth, and I needed to listen more carefully in Ableton.

[TLW]

Psychoacoustics can throw all kinds of weird tricks at us.

It’s amazing how much difference a pretty small change in volume can make our brains think the louder version of a track sounds “better” than the quieter one. A trick used by less than scrupulous hi-fi/audio visual tech sales people is to demonstrate the same track through different gear and when they get to the thing they want to sell you they nudge the volume up a bit. 2 or 3dB is usually enough.

As for comparing exported tracks, rather than loading instruments into a track I suggest a better approach is to use an audio sample with zero effects, plugins or processing used. A sine wave or white/pink noise is as good as anything. Export it twice at 32 bit depth with no dither applied. Import both exports into an empty project then flip the phase on one of them. The result should be silence or so close to it that any differences are negligible, and the best way to check for that is to export the null test again and examine the wave form visually under high zoom levels or use a sensitive frequency/loudness meter on the master track. That way any inconsistencies in the audio interface, driver or playback system such as self-induced noise from amplifiers or even from within your own ears or body won’t interfere with the comparison.

[Nokatus]

at the very least, they should trigger (initialize the phase) at render start time, even when retrigger is disabled.

No, they really shouldn't !

A separate "deterministic mode" which syncs in this manner and (in the case of random elements) exposes the seeds used in the algorithm and so on... is a nice idea. But it's definitely a separate thing and aesthetic. Free running LFOs, and so on, should absolutely never ever trigger in a repeatable and uniform fashion on render, by default.

If they do so by design, it should be a separate mode and option. (That's how you use the word "should" truthfully ).

[orin]

Some people like randomness and chaos. As an engineer, I prefer order and determinism. Starting from determinism, you have sufficient control to emulate chaos and even play with variations. Starting from chaos, your only option is to keep rolling the dice.

sNokatus, er I mean katusNo, er I mean Nokatus, you must be lucky!

[TLW]

Interesting that you mentioned dice.

When rolling dice the outcome is not chaos. Chaos implies a completely unpredictable outcome and a complete lack of constraint on that outcome.

Take three dice. Every time they are rolled the exact outcome of the roll isn’t known until they come to rest but the possible range of outcomes is constrained. It must be a positive integer between 2 and 18. The probability of particular results within that range is also constrained, some outcomes being more probable than others. There is randomness but it is constrained randomness. Not chaos.

If I play a note on an analogue synth and have a free-running LFO affecting it I don’t know at what exact part of the LFO waveform’s phase that LFO will be. But I do know range of effects it has is constrained and what those constraints imply for the result. What’s more the VCO producing the note in the first place may well not be perfectly stable either, but it is still musically useful and people generally seem to like the result.

When I do want a fixed pre-determined LFO phase then I sync the LFO to a clock source, but even then the nature of the instrument means each time the LFO cycles the waveform will be slightly different.

Similarly, it’s unlikely a guitarist or a strings, woodwind or brass player will hit a note with exactly the same frequency, timbre and volume envelope every time they play it. A great deal of what goes into making music is about more or less tightly constrained randomness.

Which is probably more than enough off-topic philosophy for one day

[orin]

I appreciate the distinction, but I'm using 'chaos' as short for "random and out of control". If the LFO is applied subtly then its effects aren't very noticeable or unmusical, but if the LFO is modulating very noticeable effects then phase control becomes more important.

Physical (analog) events are never reproduced perfectly, but software (digital) events CAN (and I argue, should) be repeated exactly: byte for byte, sample for sample. Centuries of work by mathematicians and computer scientists have gone into making computations consistent and predictable -- all we have to do is embrace the gift in software design. It's easy when you see it.

[TLW]

It really depends on what you’re doing at the time, I think.

For the creation side of things some constrained randomness can be a very useful tool, e.g. generative electronic music.

For something like a linear phase eq on the other hand, I agree consistency is important and bounces from a frozen/flattened track should always be identical.

[yur2die4]

If you want consistency, record it. Then it’ll always play back the same. But the generator on its own is intended to have a personality, it is intended to be ‘caught’. This also gives those devices certain characteristics when doubled. Otherwise they’d just full blown add together into the same tone but double volume. I doubt developers want to spend the time making a ‘switch’ for turning rando on and off. Still, I doubt Any DAW can render random devices consistently.