Speech To Phrase audio ERROR

Configuration Voice Assistant
,
1

Using a voice assistant yellow and voice preview edition I can often get usable results.

But sometimes it just sits there flashing same the voice debug shows no information.

I turned on debug logging on the speech to phrase add-on and saw the log in there showing things like this:

ERROR (online2-cli-nnet3-decode-faster[5.5]:GetLattice():online2/online-nnet3-decoding.cc:69) You cannot get a lattice if you decoded no frames.
kaldi::KaldiFatalError[11:20:17] INFO: Service exited with code 256 (by signal 9)

A quick Google seems to imply it’s not receiving the audio it’s expecting to even start the phrase processing.

Anyone have thoughts on what might be going on here?

2

Did you get this figured out? Currently trying to set up an atom echo and am running into the same issue, except every time.

3

I didn’t I’m afraid. I feel like it did improve after some arbitrary poking around but nothing with reproducible steps. I’ve since switched back to the standard voice assistant but running in containers on an old laptop.

4

So my understanding is that it did not get audio to decode.

I have been listening to my voice commands captured by voice device and audio is EXTREMLY low. at full volume I am barely able to hear my own speech in background. Conversly the wake “beep” is very loud and clear. I am thinking input volume is low so I plan to try and adjust mic input.

I have not compared audio from HAV vs Echo vs the waveshare devices I have. Currently I only listen to audio from the waveshare devices so this may need to be checked as some point.

EDIT
as an FYI, I mounted the debug voice inputs files to a media folder so I can play them back. The files are saved with random folder/file names that seem to follow no logic so I may create a folder watcher to give some indication of file creation time. If you mount this via samba or something it may be easier or maybe use the folder watcher to move it to new file that has logical name yy-mm–dd–hh-mm-ss.wav or whatnot

to save voice command input add below to HA Config

assist_pipeline:
   debug_recording_dir: /share/assist_pipeline
1 Like