I have a pi 3B+ with respeaker 2 and alsa, no pulseaudio.
Rhasspy (satellite) runs as docker, squeezelite is installed on host.
Speech comes from the base rhasspy.
I want to hear rhasspy output while squeezelite is playing music.
Can i use this? Do i need rhasspy-speakers-cli-hermes?
For Home Assistant is an integration available to play TTS using LogitechMediaServer. But it works with the TTS integration.
I did not found a way to use rhasspy as TTS for Home Assistant other then by Hermes/MQTT.
Hi!
Using the setup like i menioned in my other topic you linked, I was able to get both speach and music simultaenously. I then had to use an automation to mute the audio when the rhasspy wake word was detected. I was never able to get the rhasspy docker to work with squeezelite.
Unfortuantely, due to the music playing the wake word very rarely worked. (And i mean rarely, i felt like a fool shouting “SNOW BOY” over anything but the quietest music playing).
I pivoted to using a physical button connected to a GPIO to activate rhasspy. And then eventually stoppd using rhasspy and used physical buttons for common music commands and a tablet for the Home Assistant GUI. Rhasspy is only as good as the hardware out there and the respeaker devices a tried were not replacement for a physical interface.
Haven’t had rhasspy set up for a while guys and my memory is rather dim but I would hazard a guess it could be to do with a sampling rate problem where many setup using hardware direct as in hw:0 which often is asking for trouble as if plughw:0 was used then there is a layer that will automatically resample and convert.
Snowboy never tried as I am not really sure why commercial cast off’s past sell buy have been adopted apart from they where there but there is a reason its no longer commercial as it was an early KW system and now not very good.
Hardware wise a mic is a mic its the underlying software that we are absent of even though you can make much better noise tolerant KWS than you have.
Pico Porcupine is prob the best almost non commercial as they do give totally free models to go at its just not open source.
The respeaker is as good a mic as any as my bugbear is it doesn’t matter at all and any more than 1 mic is totally pointless if it is not beeing processed by hardware or software dsp but there is no link up with the rhasspy system anyway.
You have an open mic with a KWS that is not very noise tolerant and to rub it in you have extra mics for no purpose at all apart from adding slight phase distortion due to the distance apart.
Your missing a whole load of software that I have tried in vain to raise many times that it is missing and lacking from the framework for such a long time and haven’t got the skills myself as it needs a modicum of C to do real time DSP as python really sucks so bad its pretty pointless trying, but obviously we only have a single good programmer who was Michael but only python based creating this initial audio processing hole.
The audio out is prob just some config somewhere but exactly what and where I am slightly out of the loop as got to the stage you guys are and thought yeah I will keep with the Google Nest Audio for now.
I am back playing with tensorflow as there is definately more dataset avail now and a KWS is not hard to accomplish but the interests of others lies elsewhere.
You might be interested in my approach with Jaco-MusicMaster. The setup works with docker and you can play songs from local file system or via youtube. To improve voice understanding, the skill automatically turns down the volume after wakeword detections, similar as @Gregory_Wilson’s concept.
Even though the skill is mainly intended for Jaco, you might be able to reuse some setup steps for a Rhasspy application.