Force offline recognition in Input Speech (Experimental)

Post your feature requets for new triggers, conditions, actions and other improvements.

Moderator: Martin

Locked
User avatar
Desmanto
Posts: 2709
Joined: 21 Jul 2017 17:50

Force offline recognition in Input Speech (Experimental)

Post by Desmanto » 22 Jan 2018 17:25

Hi Martin,

I am back again while creating my speech flow. I have finished until the part to log the new recognized keyword. When I test the flow, the Input Speech recognize my wording, but it stuck at that element for about 20-30 seconds before finally continue to the next element. The recognized words are fine, but it just takes too long to finish the input speech after I have stopped speaking.

I have experienced this several times, it is intermittent. But I finally understand the root cause. The last time I test it, I am connected to the wifi, but the internet is down (the main modem is not online). So it seems the input speech try to connect to google server first to use network recognition and after timeout (about 20-30 seconds), then it will use offline recognition. I tried to disable the wifi, and there is no delay anymore. As soon as I stop speaking, the execution directly continue to the next element. So the culprit is the network recognition. Online recognition works better, but in my case offline also works quite good as well, since I have downloaded the offline pack in google app.

So if i can just force the recognition to be offline, something like "Force offline recognition", then I will never face the issue of this timeout anymore. Some of the time, even though the wifi or mobile data is working fine, approx 1 out 20, there will be this delay (20-30 seconds). But when I turned off wifi and mobile data, 50 tests, no delay at all. I would prefer the best perfomance then.

If it is not possible to force offline usage, maybe at least let us have timeout option. So after certain seconds passed, the input speech will be stopped no matter what.

For current workaround, the easiest one is to disable wifi and mobile data while I am running the flow. But maybe I should use AFWall+ or netguard to disable the internet for a while and enable it back after a delay of time.

Regards,
Desmanto
Index of Automagic useful thread List of my other useful posts (and others')
Xiaomi Redmi Note 5 (whyred), AOSP Extended v6.7 build 20200310 Official, Android Pie 9.0, Rooted.

User avatar
Martin
Posts: 4468
Joined: 09 Nov 2012 14:23

Re: Force offline recognition in Input Speech (Experimental)

Post by Martin » 23 Jan 2018 09:27

Hi Desmanto,

It seems that an app can ask the system to prefer offline speech recognition (Android 6+).
I'll add a configuration to the input speed action for the next update.

Regards,
Martin

User avatar
Desmanto
Posts: 2709
Joined: 21 Jul 2017 17:50

Re: Force offline recognition in Input Speech (Experimental)

Post by Desmanto » 24 Jan 2018 05:39

Thank you Martin. Hopefully it will still works with LP 5.1, since I just got update from google app, which allow to Google Assistant as well.
Index of Automagic useful thread List of my other useful posts (and others')
Xiaomi Redmi Note 5 (whyred), AOSP Extended v6.7 build 20200310 Official, Android Pie 9.0, Rooted.

Locked