I am able to repro this issue, and I do see it not repro on IE or Safari. But 1)I tracked back to M40 and M41, the behavior is the same, so it is not a regression. 2) Chrome never freeze all the time. It only pause a short time (<1 second) at the first sound, then all the later time it behaves correctly. I agree it is a bug, but based on the above two observation, it is either a low priority or won't fix.
Dale, what do you think? I am not sure how important the construct 2 is.

Looks like you're playing the sound w/ WebAudio? I don't see any html5 players being created.

Construct 2 is an html5 game engine. I will check about web audio and get back to you. Thanks.

Construct 2 plays all sound effects through web audio, but the first time Construct 2 plays a positioned sound, it creates a panner node. It seems in Chrome only, the creation of the first panner node causes audio to cut out momentarily. I assume it's lazy-loading the spatial audio engine?

I have not had a chance to try this yet, but Chrome does lazily load the HRTF database when the panner is created.  However, this the default panner type is equalpower, not HRTF.  Are you using HRTF?

That being said, the panner should probably be a pass through if the HRTF database isn't loaded.

From the code, I see that the HRTF panner is being created. The HRTF database is loaded when the first panner is created.  This does take a little time, but on reasonably fast machines, it's not noticeable. (On slow Android machines it was very noticeable.)

The best solution right now is to create the panner node as soon as possible and set the model to HRTF right away.  And then to wait a little while before starting.

This is caused by a locking issue in the HRTF database loader.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0494c3815a7eebceb73234e8da9e368e97fc1bfb

commit 0494c3815a7eebceb73234e8da9e368e97fc1bfb
Author: rtoy <rtoy@chromium.org>
Date: Tue Jul 19 21:44:22 2016

Simplify locking in HRTFDatabaseLoader

The locks in loadTask() and loadAsynchronously() were blocking each
other for about 100 msec or more, causing the jank in the bug report.

Only the lock is loadTask() is needed because that's what writes the
m_hrtfDatabase pointer.  loadAsynchronously() doesn't need a lock because it
doesn't need to read m_hrtfDatabase.

Also simplified isLoaded() and database() which basically do the same
thing except for the return type.

database() (and, indirectly, isLoaded()) also gets a tryLock because
it is called from the audio thread which can't block.

BUG= 602668 
TEST=No jank in test in 602668

Review-Url: https://codereview.chromium.org/2154473003
Cr-Commit-Position: refs/heads/master@{#406389}

[modify] https://crrev.com/0494c3815a7eebceb73234e8da9e368e97fc1bfb/third_party/WebKit/Source/platform/audio/HRTFDatabaseLoader.cpp
[modify] https://crrev.com/0494c3815a7eebceb73234e8da9e368e97fc1bfb/third_party/WebKit/Source/platform/audio/HRTFDatabaseLoader.h

Tested the same on win8.1 chrome version 54.0.2802.0 with the URL https://dl.dropboxusercontent.com/u/184657779/chrome%20bugs/audio%20position%20html5/index.html - Observed that sound is paused a short time (<1 second) at the first sound though the text "Playing Sound" is displayed sound is not heard. All the later times it behaves correctly

Please find the screencast

IE Behavior : Sound is heard for the first sound as well.

rtoy@, Could you please confirm on the fix as the issue is still seen

Deleted: Recording #10.mp4 668 KB

	Recording #10.mp4 668 KB View Download

I verified with ToT chromium build that the jank is gone. However, there is no sound played for the very first time.  I think this is related to the long-standing issue that if the HRTF database is not ready, the panner outputs silence. This should really be changed to make it a pass-through, but that's a different bug.

I will test on Win and OSX shortly.

Win10, Chrome canary 54.0.2802.0 has a very slight jank still.  More investigation needed.

OSX Canary is still 54.0.2800.0 which doesn't have this fix.

For the record, win10 chrome canary is way better than chrome stable on this test.

ToT chromium on OSX has no jank, and, unlike linux, it does play the first sound.  However, it seems truncated; it's shorter than the other sounds.  I think this is, again, the panner outputting silence until the database is loaded.

See  issue 629858  for the silence issue on the first sound.

tkonchada@ Can you clarify your comment in #c9?  Are you saying that you don't see the janky animation anymore but you also no longer hear anything?  If this is the case, it is fixed in  issue 629858 .  

While I do possibly see some animation jank it's really hard to know for sure and to know if it's not caused by other things going on on that page.  

rtoy@, I can confirm that there is no janky animation but the sound is not heard for the first time. Later it is working fine with the sound as well.

Thanks for the confirmation.  The lack of sound the first time is  issue 629858 , where a fix landed yesterday.  

Closing this (verified), based on my own tests and your tests.