Audio clipping / static / interference on rapid intervals of sound (fixed in master)

[kezzbracey]

Godot version:
3.1 alpha

OS/device including version:
Kubuntu 18.04 and Windows 10

Issue description:
Creating a rapidly firing weapon, if sounds are played at intervals less than 0.5 or 0.6 seconds there is a clipping or static sound.

Demo: https://www.dropbox.com/s/q5tx4y6272lgdae/2018-09-12%2017-16-00.mp4?dl=0

Improvement can come from modifying sound files with fade in and fade out at either end, but the issue is still present.

Demo: https://www.dropbox.com/s/hnvlu161x1t62fl/secondtest.mp4?dl=0

In the case of testing with a sine wave sound file, instead of clipping / static there is a sort of interference or modulation sound instead.

Demo: https://www.dropbox.com/s/nd2rcsrdrqxo6sd/2018-09-12%2021-19-52.mp4?dl=0

The clipping / static sound can seemingly be eliminated by setting a LowPassFilter and LowShelfFilter on a sound bus, both with a cutoff of 1000Hz.

These filters don't eliminate the modulation type of sound that comes up when using a sine testing sound.

Steps to reproduce:
Create a timer with an interval of about 0.3 seconds and have either a single sound repeat at that interval, or rotate between two different sounds.

Issue present in the editor on Linux and Windows, in builds for Linux and Windows, on 2 PCs in my house, through 3 headphone sets and two speaker sets. Also verified by another user who tested on 2 of their headphone sets.

Minimal reproduction project:
SoundTesting.zip

Tests and attempted solutions that didn't help
Before trying the filters, I also tried:

• Having an AudioStreamPlayer2D as a child of each instantiated missile (this made the effect worse)
• Having the Player character scene play a sound from a child AudioStreamPlayer or AudioStreamPlayer2D
• Creating two child AudioStreamPlayers and alternating between playing one or the other, using the same sound
• Alternating between those two child nodes, but playing two different sounds
• Creating a stand alone scene with an AudioStreamPlayer in it, and instantiating it newly and adding it as a child each time
• Outputting the sound from the two child AudioStreamPlayers to different audio buses
• All three different mix target options
• Reducing volume - tested at -2Db, -4Db, -6Db
• Tested with 9 sound files
• Modified a sound effect with fade in and fade out processing to ensure no clicking in the file itself
• Tried reducing the duration of the sound file to 0.3 seconds, but still had the static effect at intervals of 0.5 seconds or less
• Trying with different sound playback hardware
• Tested on Linux and two installations of Windows
• Tested on 2 PCs
• Tested in editor and in Linux and Windows builds (Windows builds tested on two PCs)

Extra information

I spent a few hours with the helpful folks at the Godot discord last night going through various test to try and isolate the source of the clipping / interference sounds. Our tests did lead us to believe it wasn't an issue localized to my setup alone.

Different people heard different effects from the videos and test project above. Some heard clipping / static on every sound, some heard none at all. One other person heard it as clearly as I did, and confirmed the same on two of their headphone sets.

@starry-abyss had the theory it was something to do with too many high frequency sounds at once, hence we tried the low pass / shelf filters which eliminated the clipping / static sounds.

Given the effect of the filters I'm not sure this is actually an issue as opposed to a sound management technique being required.

If we can confirm this is just about handling sound in a certain way, perhaps I can contribute to the docs with some information for people in the future who run into this problem.

[Calinou]

Try reimporting sounds with the Trim property disabled:

I've found Godot's automatic trimming to be too aggressive, resulting in part of the sound not playing.

Also, beware that Godot 3 currently does not provide polyphony (#23541), so instancing an AudioStreamPlayer for every sound played is a good idea if you need several of them to play at the same time. (Of course, don't forget to queue_free() them when they are done playing.)

[Aaron-Fleisher]

FWIW this does not happen on 3.0.6 for me. I sometimes get a random "static sound" (very rare, but it happens) at the beginning of certain sounds that are played (.ogg). Never heard that kind of static in 3.0.6. W10, 64-bit

By the way, that's a very extensive issue report, kudos.

[lawnjelly]

Godot version :
3.1 stable
OS :
Linux Mint 18.2 Cinnamon 64 bit
Issue description :
Glitches in audio when reusing AudioStreamPlayer3D node

It sounds like this may be the same issue I have been dealing with in the past couple of days. I have a workaround that works, and a possible reason for what I now believe is probably a bug rather than calling the AudioStreamPlayer incorrectly.

Minimal reproduction project:
godot_audiostreamplayer_test.zip

Full description of the problem
Rather than creating new audioplayer nodes (in my case AudioStreamPlayer3D) each time I need a sound effect, I would expect to either:

• A : Create a pool of audioplayers, select one that is free and use it to play a sound in the desired location and volume
• B : Have a permanent audioplayer attached to game objects, and use it to play sounds as needed

Both of these cases involve reuse of the node to play multiple times.

A
When using this scheme, if the AudioStreamPlayer translation and / or volume is changed prior to playing, audio glitches occur. This is not to do with the import settings / normalization / trim, it occurs with these off. It is not clipping. The glitches on my system are approx 1/20th second, probably one of the audio tiles used in a circular audio buffer.

Picture of glitch

The glitch can be seen bottom left waveform, at the start. Each sound should be playing for 1 second, and the glitch is around 0.05 second.

Audio of glitch
https://soundcloud.com/lawnjelly/godot-audio-glitch-test

B
When attached to a game object node, there are not problems from translate (as it is not set prior to calling play()), however changing volume still results in glitches.

Workaround:
By chance I found a different method of playing sounds by Calinou, which I slightly modified and solves the issue:

func PlaySound_Node(var stype, var node, var volume = 1.0):
    assert (node != null)
    var player = AudioStreamPlayer3D.new()
    node.add_child(player)
    player.stream = StreamFromSType(stype)
    player.set_max_db(-60 + (volume * 60.0))
    player.connect("finished", player, "queue_free")
    player.play()

This seems to solve the issue, and works as a workaround. However, the question is, why?
Creating new AudioStreamPlayer for each sound and attaching to the parent node and deletion on completion doesn't sound the most efficient way of playing sounds.

At a guess, I suspect the reason it works may be because creating a new AudioStreamPlayer is forcing creation and blanking of an audio buffer (or position within a buffer) whereas reuse of AudioStreamPlayer is not getting rid of previous data from playing (buffer, position within buffer, historical record of position or volume?).

I also discussed the issue here:
https://godotdevelopers.browse-tutorials.com/discussion/comment/27395

[starry-abyss]

@lawnjelly Just a quick thought: do you initialize the state of objects (i.e. make sure all internal variables are not holding leftover values) before reusing them from the pool? As this is a common caveat with pools. When the new one is created (case B), it's clean, but is so the one in case A?

[lawnjelly]

starry-abyss, the example project I posted should show exactly. I don't do anything to the AudioStreamPlayer between playing one sound and the next, just move the translation, change the volume, and call play() again after the previous sound has finished.
Is there anything else that should be done?

As a user I would assume that a sound would autostop at the end, and another play would reset everything necessary. I did actually try calling stop() before play just in case but that didn't change anything.

Calinou, the problem still occurs when trimming / normalizing is turned off during import (you can try this on the example project I posted, and reimport the wav). So I'm assuming it isn't to do with trimming / normalization.

[Calinou]

Calinou, the problem still occurs when trimming / normalizing is turned off during import (you can try this on the example project I posted, and reimport the wav). So I'm assuming it isn't to do with trimming / normalization.

I realized that just after posting the comment, which is why I deleted it 😉 Sorry for the confusion.

[jitspoe]

There's also a problem with pops in the AudioStreamPlayer3D. Just going to paste the contents of my duplicate bug here for easy reference:

Godot version: 3.1.2, 3.2 rc4

OS/device including version: Windows 8.1

Issue description:
There are a few situations where playing a sound AudioStreamPlayer3D results in pops or crackles. The easiest to reproduce has something to do with the positional logic. The surround sound logic changed between 3.1 and 3.2, so I thought maybe it was fixed, but even with the new 3.2 positional code, it still has the pops.

I replaced my character's footstep sounds with sine waves, and it became VERY clear that there was something wrong.

It looks like there are actually 2 problems: One, seen in the top channel, is an abrupt volume/amplitude change. The second, shown in the bottom channel, shows the sine wave getting out of phase. I think it's actually playing the sound again at a later time and blending it (also followed by an abrupt amplitude change).

The abrupt change seems to happen at 21ms, which, at a 48000 sample rate is pretty close to the 1024 buffer size. My guess is the first buffer somehow has bad data and is mixing/fading incorrectly.

I dug into the code for a while, but couldn't identify the exact problem. From the looks of it:

1. AudioStreamPlayer3D::play() indicates the player is active and enables the physics process.
2. AudioStreamPlayer3D::_mix_audio() is called from a separate thread. Data isn't set yet, so it fills the buffer with 0's (this is probably not great, as it likely delays the sound, but shouldn't cause a pop).
3. _calc_output_vol() called when there's a NOTIFICATION_INTERNAL_PHYSICS_PROCESS, which sets proper data.

Seems there is something wrong with the first buffer every time:

You can see here that sometimes the amplitude is scaled up, and sometimes it's scaled down. Rarely is it ever at the volume it should be (unless maybe the sound is played at a location very close to where it played previously?)

Other pops:

There are 2 other pops I'm aware of that may or may not be part of the same issue. Perhaps they should have their own bug entry once this is fixed.

• A rare, but very harsh pop that happens seemingly at random. Does not appear to be position related (I hardcoded the location to test). Can't figure out a consistent repro for this one.
• Calling stop() on a looping sound.

Steps to reproduce: Play a sound on an AudioStreamPlayer3D, then move it to a new location relative to the camera and play it again. Really obvious when you move from the left to right side of the listener (see attached project).

Minimal reproduction project: sound_pop.zip

[lawnjelly]

3. _calc_output_vol() called when there's a NOTIFICATION_INTERNAL_PHYSICS_PROCESS, which sets proper data.

This may be the problem. I had assumed before it would be something to do with the audio buffer tiles (without looking at the source code), but it may simply be the audio is only changing things on physics ticks.

Besides using physics ticks (which is kind of dodgy, because it means changing physics affects the audio) It needs to be doing interpolation for everything between the ticks, because audio needs to be continuously interpolated. I'm guessing it isn't somewhere.

Also I suspect the problem I was having with changing position and playing a AudioStreamPlayer again was because it was waiting to the next physics tick to set the new output, instead of doing this via a forced path every time play is called.

I can't look at it now, busy with other stuff, but hopefully this might provide ideas for anyone else working on the audio.

[jitspoe]

I tried a quick hack where I forced a call to update the physics-related stuff to try to ensure the data was set before the audio thread tried to process it, but that didn't seem to fix it. :(

void AudioStreamPlayer3D::play(float p_from_pos) {

	if (stream_playback.is_valid()) {
		setplay = p_from_pos;
		output_ready = false;
		set_physics_process_internal(true);
		_notification(NOTIFICATION_INTERNAL_PHYSICS_PROCESS); // HACK - force position update before audio driver tries to process this.
		active = true;
	}
}

The audio thread keeps track of the previous data and lerps between the previous and current values. This appears to work properly everywhere except for the first buffer.

[jitspoe]

In case it's helpful, here's a clip from when I got a pop while hardcoding the relative position of the listener (WARNING: TURN VOLUME DOWN): https://clips.twitch.tv/ViscousCooperativeBasenjiMoreCowbell
Another: https://clips.twitch.tv/JollyDignifiedCarabeefItsBoshyTime (you can also hear the pop when the elevator looping sound stops, which is another issue).

Unfortunately, the compression makes it almost impossible to tell what the original wave form actually did.

I think there are 3+ unique pop bugs.

[Calinou]

Unfortunately, the compression makes it almost impossible to tell what the original wave form actually did.

You could upload a short video with uncompressed audio, put it in a ZIP archive and upload it here. Or just upload a WAV file of the video's uncompressed audio in a ZIP archive 🙂

[jitspoe]

I don't have an uncompressed recording. Whenever I try to catch it in the act, I can't seem to reproduce it. I just happened to have these because I'm live streaming the whole development.

[jitspoe]

1/3 of the pops fixed with Waridley's change! That was probably the biggest one. There's still this rare pop that happens even after the change. Only happens like once a day for me. It can be really harsh, so I hope we can track that down, too. Not sure if the stop pop is still there, as I worked around that one.

[jitspoe]

@akien-mga Sadly the other pop is still pretty prevalent. Seems to happen most frequently when sounds are first played. I'm wondering if something is uninitialized (prev_outputs, maybe?). Happens both on Windows and Linux (completely different machine). Actually seems more prevalent on Linux, but maybe that was just by chance.

[jitspoe]

Ok, I've finally managed to capture a recording of the sound pop with audacity, so it's a little cleaner. Unfortunately, there's some resampling going on so it's not perfect, but at least there's no compression:

Wav file:
godot_random_pop.zip

What's crazy here is that audacity scales the volume based on my windows settings, so normally if there were a pop like that, it wouldn't peak unless I had everything at max volume. Playing the sound back and recording it with my volume set at 43%, it looks like this:

Meaning that somehow Godot is bypassing the max volume in windows and just flat out spiking as loud as the sound card and headphones will let it go. That explains why these random pops are so much worse than the other ones (that were fixed).

This really needs to get fixed. It could blow a speaker or damage somebody's hearing.

[jitspoe]

Leaning toward this being an uninitialized value. I changed the AudioFrame() constructor to set l and r to like 99999, and it seems to happen more often now, but I can't find a consistent repro case for it. Sometimes it seems to happen when changing direction. Sometimes it happens like every footstep. Wav file where it was happening a lot:
godot_random_pop3.zip

It seems to always pop upward and stay flat for a period of time, so maybe it's not just a weird volume scale.

Ok, I've recompiled with the AudioFrame() constructor defaulting l and r to 0, and I can't seem to get the pop to happen anymore, but it is somewhat random. I don't think this is the solution. It likely masks the problem and introduces another (more subtle) problem where the audio is too quiet and fades in.

[jitspoe]

Yep, definitely an uninitialized value somewhere. I had the constructor set l = 99999 and r = 0, and now the popping only happens in the left channel. What's weird is that it doesn't happen EVERY time a sound is played, and I can't seem to make a simple repro case scene. It's possible it's a threading issue that only happens when enough stuff is happening simultaneously.

Side note: The volume for 2 channels vs 6 (surround) channels is different. If I have my sound configured to use 6 channels then output for headphones, it's considerably louder than if I use 2 channels. Also, the panning seems really harsh. If the character is slightly to the left of the listener, there's almost nothing coming out of the right channel when using 2 channels. Sounds more natural with 6.

[jitspoe]

Slowly honing in on this. Seems the problem is in the buffer coming from the stream playback sample is bad, now, not the volume. It might be related to changing the pitch of an ogg file. I was randomizing pitches, which would explain why the problem happens randomly.

Here's the buffer (note left channel is large and right channel is 0 all the way up to just before sample 256):

[jitspoe]

Just posting all this info bit by bit because I haven't had time to sit down and focus on it, but the problem appears to be with the buffer in AudioStreamPlaybackResampled::mix()

internal_buffer has uninitialized data.
INTERNAL_BUFFER_LEN is 256, which explains why the data clears up after 256. Now, initializing the data to 0 would fix the pop, but there's another issue where sounds sometimes don't play (every once in a while, I'll get a silent footstep, for example) which might be related. Need to track down why the data isn't getting set before mix() is called.

Edit: Seems it only happens with ogg vorbis. I have a consistent repro project now (might only happen consistently if you set the default AudioFrame constructor to init l and r to large values):
sound_pop2.zip

Edit2: Side note, it would be nice if we could load ogg vorbis files into memory so we could keep game file sizes down without the perf hit of streaming them.

Edit3: I forgot to set the ogg to not loop, so there's ANOTHER popping issue when a looping sound is interrupted which should also probably be addressed. If you disable loop on the project and reimport, the pop only happens every 1 in 4 runs or so.

[ellenhp]

In 4.0 you should not get pops under any circumstances. In 3.x you still get pops during these 3 things:

• change audio buses on a player while a sound is playing,
• queue_free a player while it's playing
• change the stream of an audio stream player while it's playing

But I don't think I'm smart enough to fix them. It would be an enormous and incredibly error prone change to make. It took a pretty substantial rewrite of the audio system to fix those pops for 4.0 and it's just too big of a change to backport. It's usually easy to avoid doing those 3 things though. Definitely let me know if you're doing anything other than those 3 things and seeing pops though because those are the only cases I'm aware of right now.

[hhyyrylainen]

If it's coming in 4.0, I think we can live with that. There's already a few other big fixes in 4.0 that we are waiting for already.

[Calinou]

Reopening, as the bug still exists in 3.x and we should try to get it fixed there, or at least document how to avoid it.

[sketchyfun]

Just attaching the project from my most recent ticket about this, so that it can be easily tested. Project shows the bug occuring when the sound is rapidly played. Doesn't occur when a regular (non 2D/3D) AudioStreamPlayer node is used

audio_pop.zip

[wizbane]

Here's a demo project for 3.4.3 that shows popping/clicking occurring when an AutoLoad AudioStreamPlayer is playing, and more AudioStreamPlayers get added to the scene. It occurs when they are added to the scene, but not playing, and -80db, stream paused. This is primarily an issue when streams with OGG files get added, occuring when even 1-2 are added in rapid succession. It does occur when 500 or so WAV or MP3 streams get added too, but not at more reasonable numbers.

audio_pop_demo.zip

[ellenhp]

I'm not entirely sure @wizbane's issue is the same as what was originally reported in this bug, but regarding @wizbane's issue, I think it's just caused by the AudioStreamPlayer nodes holding a global audio server lock during most of the execution of the set_stream method. Stream playback instancing and initialization like that prevent the audio thread from mixing, which is bad news. It gave me bad vibes during the 4.0 audio mixing rewrite so I removed the need for locking, but solving this in 3.x is probably still worth doing.

[ellenhp]

If anyone has a repro on this bug and would be willing to test it against #59413 that would be wonderful! I tested the project in #22016 (comment) already but it seems like there might be other ways of causing this to happen?

[akien-mga]

Fixed by #59413.

[sketchyfun]

I've just tried applying the changes from #59413 but when I run the project I attached in my comment above I'm still getting popping noises, so I don't believe the issue has been completely fixed. Can you try the file I attached to my previous comment and see if it's fixed for you?

[ellenhp]

That repro project still displays the pop. Seems like the cross-fade is broken. The 3.x audio system is really tricky to work on but I bet it would be possible to move the non-spatial player crossfade logic into the spatial players. No promises though because my availability for Godot work has been really inconsistent lately.

[ellenhp]

As a workaround, you could try using multiple audio players and allowing playback for each to expire normally when the audio stream ends.

[sketchyfun]

As a workaround, you could try using multiple audio players and allowing playback for each to expire normally when the audio stream ends.

This happens in many places throughout my game, so I don't think multiple audio players would be a suitable solution for me. I'll have a dig through the regular audio player code and see if I can figure out what can be transplanted into the spatial audio player to fix my problem

[ronyeh]

Hi!

If any devs on the Godot Audio system need me to help with testing or bug repros, please @ or message me. I've been running into issues with my app regarding rapidly repeating sound effects (and pitch shifting), so I am very much interested in seeing the audio system improved. I'm happy to help out in any way that I can.

Thanks for all your hard work. Godot is awesome. :-)

[elvisish]

Is there any workaround for this in 3.x? Even using 4+ audio players and iterating through an array to assign which one to play doesn’t work for some reason.

[elvisish]

Is this fixing the problem with zero-crossings causing the pops? I opened 5697 but it was closed as it was suggested to be a duplicate of this, but I don't think so. I think Godot needs to automatically add a micro fade out at the end of sounds when it stops playing them to avoid pops/clicks on zero-crossings. This will happen any time a sound is stopped and it lands on a zero crossing, be it destroying the audio player, stopping the stream, changing the stream or seeking. It's not a bug really, it's just a natural mathematical inconvenience that should be accounted for in the audio engine.