Mycroft not responding during audio output?

Category · April 21, 2019, 10:20pm

I am currently running Picroft on my Raspberry Pi 3B+, using a Logitech EasyCall USB Speakerphone for the interface.

I’ve installed pulseaudio, and set the sink/source correctly for normal function. Mycroft will pick up the wake word, and answer any queries.

However, the “Stop” skill is only working from the command line.

For example, I ask for the news from BBC, and Mycroft will start playing the audio stream - but will not respond to the wake word, or me saying stop, until the audio has finished. If I need to stop Mycroft earlier, I need to go into the command line to issue the stop command. This is the same during the output of any skill.

I have loaded the echo cancellation module in pulse audio, but that has made no difference. Any ideas? This is currently the biggest of the 2 issues I am facing with Mycroft, and have been banging my head against it for several hours now!!!

gez-mycroft · April 23, 2019, 6:00am

Hi there,

Just want to make sure that when attempting to say “stop” to Mycroft, are you first using the wake word, eg “Hey Mycroft, stop”?

Does the behaviour change depending on how loud the audio is? Is the microphone unable to distinguish your call from the surrounding audio? Or does this happen regardless of the audio levels, or if you are using headphones?

The difficulty in breaking into an audio stream is one of the biggest limitations of standard audio devices being used as a voice assistant. This unfortunately extends to our Mark 1, as they aren’t performing their own echo cancellation, beam-forming, noise suppression etc.

It is something that will be greatly improved with the Mark II, and why we’ve selected hardware that has been specifically designed and built to natively support these features.

j1nx · April 23, 2019, 12:21pm

Isn’t the listener by default muted on playback because of those Mark-1 troubles?

github.com

MycroftAI/mycroft-core/blob/dev/mycroft/configuration/mycroft.conf#L173


      
          // 'record_wake_words' and/or 'save_utterances' are set to 'true'.
          // WARNING: Make sure that user 'mycroft' has write-access on the
          // directory!
          // "save_path": "/tmp",
          // Set 'record_wake_words' to save a copy of wake word triggers
          // as .wav files under: /'save_path'/mycroft_wake_words
          "record_wake_words": false,
          // Set 'save_utterances' to save each sentence sent to STT -- by default
          // they are only kept briefly in-memory.  This can be useful for for
          // debugging or other custom purposes.  Recordings are saved
          // under: /'save_path'/mycroft_utterances/<TIMESTAMP>.wav
          "save_utterances": false,
          "wake_word_upload": {
            "disable": false,
            "url": "https://training.mycroft.ai/precise/upload"
          },
          
          // Override as SYSTEM or USER to select a specific microphone input instead of
          // the PortAudio default input.
          //   "device_name": "somename",  // can be regex pattern or substring
          //       or

Putting that section in your custom mycroft.conf and enabling it should do the tick I believe.

“listener”: {
“mute_during_output”: false
}

Category · April 25, 2019, 8:30pm

@gez-mycroft - Yep, i have been trying to use the wake word before saying stop, but with no luck. No level of audio seems to listen to me after that. I am running with a USB speakerphone on a RPi, so may have to see if I can tweak pulseaudio somewhat…

I will find some headphones and see if it is just interference from the speaker itself though

@j1nx - Thanks for the tip, I’ll chuck that in mycroft.conf and see if it helps.

I will also play with the threshold values on my wakeword, see if it pick anything up when it is “looser”, will report back