Voice response is incomplete

[zeusshuang]

Hi AVS team,

SDK 1.8 , ARM based OpenWRT embedded linux.

We got many reported that the voice response is incomplete or cut off.
Sometime it happens while barge-in during playing Music, sometimes it's from normal Q&A case.
Here I attached a log from barge-in case.
While playing music, say: Alexa, Thumbs down. It response "okay", but from APP History, it should respond "okay rating saved".

19541-cc1d849774154846b9c75a02703bb28d.txt

[kclchan]

Hi @zeusshuang,

Would you be able to point out the line number in the .txt file where the voice response is cut off?

Thanks,
Kenneth

[zeusshuang]

Hi @kclchan ,

By tester, the timestamp is about 2018-07-09 03:42, where you will see a "KeywordObserver" log, it's barge-in, the response cut off should happens after this barge-in

[kclchan]

Hi @zeusshuang,

Thanks for providing the timestamp, it's extremely helpful. I am looking at the logs now, and will get back to you ASAP once we find anything.

Thanks,
Kenneth

[kclchan]

Hi @zeusshuang,

Based on the logs:

018-07-09 03:42:35.665 [ b] 0 DialogUXStateAggregator:transitionFromThinkingTimedOut
2018-07-09 03:42:35.798 [ 3] 0 MediaPlayer:callingOnTags
2018-07-09 03:42:35.798 [ b] 0 DialogUXStateAggregator:setState:from=THINKING,to=IDLE
2018-07-09 03:42:35.799 [ 3] 0 MediaPlayer:callingOnPlaybackStarted:currentId=260
2018-07-09 03:42:35.799 [ 3] C SpeechSynthesizer:onPlaybackStarted:METRICS::Location=SpeechSynthesizer Receive,NAME=SpeechStarted

The speech is cut off due to getting transitionFromThinkingTimedOut before the SpeechSynthesizer start speaking. There appears to be a race condition and we have created an internal ticket to track and fix this issue ASAP.

Thanks again for letting know about this issue and providing the logs for it.

Thanks,
Kenneth

[zeusshuang]

Many thanks for your comments !

[jpatu001]

Hi All,

Is this issue resolved? I may have a similar issue. When cancelling a timer/ alarm, the response I get is "Timer for 10..." then gets cut off instead of "Timer for 10 seconds cancelled". My logs show that the transition I have when cancelling is Listening->Thinking->Idle->Speaking->Idle. This is on sdk v1.8.1 embedded linux..

Thanks,
jpatu001

[jan4984]

yes I have this problem in some arm Linux board . is this a issue of gstreamer? https://lists.freedesktop.org/archives/gstreamer-devel/2015-March/052131.html

please any one know gstreamer can help fix this for avs-device-sdk?

[jan4984]

I found that the version 1.5 sdk 's sleep works in 1.7.1. not know why amazon remove it.

MediaPlayer.cpp:

case GST_MESSAGE_EOS:
    std::this_thread::gp_for(std::chrono::milliseconds{500});

[scotthea-amazon]

Hello zeusshuang,

Release 1.10 includes a change to DialogUXStateAggregator to address this problem. If you can, please let us know whether or not this fixes the problem for you.

Thank you,
-SWH

[zeusshuang]

Hi @scotthea-amazon ,

Thanks!! I will merge this release and try it.

[KrisBird]

I just installed 1.10 on a fresh Raspberry Pi3 Stretch and the response is cut in the start.

[kclchan]

Hi @KrisBird, is your audio being output via HDMI? If so, could you try using a headphone jack instead?

[kjkh]

Hi @zeusshuang,

Are you still experiencing this issue after updating?

@KrisBird, are you still seeing the issue after switching audio output to the headphone jack?

[zeusshuang]

Hi @kjkh ,
so far I did not see this problem in SDK1.11.

[KrisBird]

I never found a reason for it not working and I ended up reinstalling. After that no problems :-)

[visayamv]

Looks like this issue is no longer existent. I am closing this issue, please reopen if you need more assistance regarding this.