Using the wake word results in a less accurate result

I have just started with Rhasspy and I am pretty impressed :slight_smile:

As I posted here:
I do have one problem though, when in the Speech tab clicking on Hold to Record , my intents are almost always recognized. However, when I use the Porcupine wake word, it is far less accurate. Does it somehow use a different STT engine then? Or is there some other difference?

@synesthesiam, replied:

It goes through the same STT engine, so this almost certainly indicates a problem with the microphone configuration. That’s good news, since it should work great once you get it fixed! If you could post to the Rhasspy community site or open a Github issue with the details of your set up, I’d be happy to help.
Most commonly, switching between PyAudio and arecord for microphone input helps. In other cases, you’ll need to adjust your microphone volume or edit your asound.conf file.

Do you mean that I have to set it to arecord and then change it back to PyAudio?

About editting the asound.conf, do you mean that when I click Hold to Record or use the wake word, that the volume is different?

1 Like

I mean switch to the other system (from whichever one you’re currently using).

When you record in your browser, Rhasspy is using the microphone data from it (Chrome, Firefox) instead of directly from the microphone. It’s possible that the two are configured differently.

One helpful thing to do when debugging is to run this command on the computer that will run Rhasspy:

arecord -r 16000 -f S16_LE -c 1 > /path/to/my.wav

Where you replace /path/to/my.wav with a place to put the WAV file. Say thing into the microphone, then hit CTRL + C to stop it. Next, play the WAV file and see how it sounds (you can use aplay /path/to/my.wav for this).

I am 100% sure that I am not using my laptop’s mic via my browser. So I am using the same one as when I am using the wake word. My interpretation of your answer is that there isn’t a difference in clicking the Hold to Record or using the wake word. Is that true?

Unfortunately, I do not have CLI access because I run Rhasspy as a Hass.io add-on on a RPi 4.

Do you have hassos or raspbian as a host system?

I have HassOS on my Pi.

Please check this thread
It will give you the way how to run commands (arecord) inside rhasspy docker container

And how to change asound.conf if required https://github.com/synesthesiam/rhasspy/issues/119

1 Like