Delay in speech recog. after update to One UI 7

[kurtkaufman]

Hello,

In this flow-

https://llamalab.com/automate/community/flows/41998

-the onset of speech recognition has always been fairly quick, within about a second of flow starting. Since updating phone (Samsung S23) to One UI 7, there is a substantial delay, even as much as 6 seconds. As you can see, a short check for existing calendar preferences is performed first, so I don't know where the bottleneck is. 

I have not changed Automate app settings per the OS; I don't believe its background processes are being killed, but it's possible Samsung have moved things around and I don't know where to look.

Any help appreciated.

Thanks, Kurt

[Henrik "The Developer" Lindqvist]

Ensure the delay is in the Speech recognition block, e.g. using a simple test flow with just that block.

Try with and without the Offline option enabled.

Is the delay before or after the prompt beep, or after you've finished speaking?

[kurtkaufman]

I set up a simple speech recognition block with the output as a variable to a toast message. No delay.

Tested with the online recognition allowed/excluded. Did not seem to make a difference. The difficult thing is that the problem is intermittent, but seems more likely to happen when Automate has been inactive for several minutes. In case it makes any difference, the Calendar flow is always run from a Home screen shortcut.

The sudden onset of the problem after update to One UI 7 is suspicious.

[kurtkaufman]

The delay is always before the prompt beep and before the toast message prompting user to speak appears.

[Henrik "The Developer" Lindqvist]

So there's no delay in your test flow? If so then it's likely some other block causing the delay.

Ensure to disable any "power save" feature, see here. Try using an "Timer accuracy workaround" in Automate setting, e.g. using the "Alarm clock" options.

[Kurt Kaufman]

Henrik, I made some changes at the OS level (since One UI 7, they have changed wording and placement yet again), and we'll see if this makes any difference. If not I will look into the timer accuracy workaround that you mentioned.
Thanks for your assistance.

--
You received this message because you are subscribed to a topic in the Google Groups "Automate for Android" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/automate-user/z2mIodt0WNE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to automate-use...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/automate-user/be1eecb8-ee70-4ef6-bd4b-f1457226ba29n%40googlegroups.com.

[Kurt Kaufman]

Extract from log. 

7036-7043

The speech recognition was sufficiently delayed at this point (although the initial toast message was not), that by the time I realized it had not functioned and the speech prompt beep was heard, I had already spoken. I decided to let it fail just so I could locate it in the log more easily.

You can compare it to running the flow the following time, 7044-7052, where a string was successfully sent to the calendar. I don't remember there being a delayed that time.

I hope this is of some use.

On Tue, May 6, 2025, 12:35 Henrik "The Developer" Lindqvist <henrik.l...@gmail.com> wrote:

--

[Henrik "The Developer" Lindqvist]

Sadly the log don't tell where the delay occur. But your flow seems to be manually started, not it response to some calendar event, which would be affected by any "timer accuracy" issue.

The delay is likely in the speech recognition engine. It's not the first report of such issues.

[Kurt Kaufman]

Thank you for looking into this for me. Since this is largely for my own use, I can deal with a delay, and not speak until I've heard the audible prompt. Hopefully the delay will be fixed with an OS update at some point.

To view this discussion visit https://groups.google.com/d/msgid/automate-user/ef143b44-0aaa-4764-ba55-9178f7e019b5n%40googlegroups.com.

[Kurt Kaufman]

I deleted the initial toast block, the prompt to indicate to the user that they should speak. Now the speech recognition begins almost immediately as evidenced by the audible prompt. I did have the toast block set to proceed immediately, so I'm not sure how that might cause the problem. Any ideas?

To view this discussion visit https://groups.google.com/d/msgid/automate-user/ef143b44-0aaa-4764-ba55-9178f7e019b5n%40googlegroups.com.

[Henrik "The Developer" Lindqvist]

It should be proceed "Immediately" by default, using "When hidden" should take a few seconds since while the toast is showing, using "When shown" should be quick unless there are other toasts already showing.

[kurtkaufman]

I tried the flow on a Samsung A52 5G, an S9, and a newly-acquired S25. No speech recognition delay problems with any of them. The problem would appear to be something in the SamsungS23 setup. I'll eventually reset the S23, and I anticipate the problem to be resolved there as well.  Thanks again for your help, Henrik.

Kurt

[kurtkaufman]

I am happy to report that after updating to One UI 8 (on Samsung S25, but likely for all applicable models), the delay has completely disappeared. The flow now functions as expected.