Echo show 5 2 gen, TTS loop/repeat

[alexreddy78]

Dear Developer,

Just noticed that on the latest app and integration version, TTS has a loop. For example: TTS is consisting of 2 sentences with 20 words. After start service TTS say, Echo Show 5 starts speaking, but in a few seconds repeat few words from the beginning and continue. The larger amount of text, the higher the number of repeated words or phrases from the beginning. I have no issues with HA Voice PE with same text and TTS.

[alexreddy78]

Dear Mark,

Any clue? By the way it’s happens only with TTS service. If I use voice trigger and asking something, all ok. How can I debug TTS?

[msp1974]

Hi. Been working on some other features but will look at this over weekend.

[alexreddy78]

More input for you, Mark. TTS Is on Omni Voice project. I played with keys for the Omni server and got flw: they have num of steps (1-32) for quality synthesis and if number is less than 10, TTS is OK, if more TTS start to repeat first sentence and then continue. However, on Voice HA PE nothing like this with same TTS and high number of steps. I think VACA handled TTS in another way and can may be improved.

[alexreddy78]

And some comments from guys who support Omni voice Wyoming server:
Esp in this scenario voices the incoming chunks and waits for the next ones.

The client on the Py library decided not to activate streaming, but waited for the full sound and only then started playback.

Step to reproduce bug:
use a slow tts that works slower than real time.

[msp1974]

Ok, help me out a little more here. So this is not built in tts but a Hass integration. Can you point me at the install so i can set this up to try.

Not sure of the context of your last comment. Vaca doesnt use ESP but Havpe does. Wyoming (what VACA uses) also sends the response before it is fully formed in the same way that ESP does and therefore does not have to wait for a long response before starting playback. Tell me more about relevance of this and what i need to do to recreate this issue.

Thanks

[alexreddy78]

Dear Mark,
Pls https://github.com/mitrokun/wyoming_omnivoice
But you need either Mac mini m4 or PC with NVIDIA GPU RTX min 4-6 GB

Regards

[alexreddy78]

Dear Mark,
Any luck here?

[alexreddy78]

Dear Mark,
Here is more detailed and professional explanation (same on AVA/VACA)
brownard/Ava#86 (comment)

[alexreddy78]

Dear Mark,

Sorry to bother you again, any luck on this?

Regards

[msp1974]

Not yet. I read up on the Ava repo link you sent me but this seems that it is a much more complicated issue as it is maybe an inherent Android media player issue. I also have no way to run omnivoice at present to even try and resolve.

Can i also confirm though that the basis of this issue is when the omnivoice is producing the tts stream too slowly and that if it didnt repeat then it would have big gaps in the audio? This was my takeaway from the ava repo discussion..is this correct interpretation?

[alexreddy78]

For sure there is something wrong with Android player. There will be no gaps, as it was tested on HA Voice PE. No loop/repeats. HA Voice PE handle this in another way than VACA. On your previous release 0.10.0 (before tts streaming implementation) there is no such issue with loop/repeats, but sadly no streaming :(. Media starts TTS only after whole generation text finished. Hope, that someday issue will be solved :)

[alexreddy78]

There is new input on AVA discussion. Would be nice, if you take a look. May be some new points to implement on VACA

[alexreddy78]

Hi, Mark!
Hope you well!

AVA developer implemented fix for it. Works excellent! Hope your fix will be soon :))))

BRGDS