News about Voice, LVA and more

[florian-asche]

We will update this thread with news and plans what we plan to do and what is currently going on in the world of voice.

[florian-asche]

Packaging and deployment

• LVA should become installable as a simple .deb package. It also serves as a small proof of concept: can we get by without Docker, to save resources on a Raspberry Pi Zero 2 W?
• We'd also build a .deb for the peripherals and add a Sendspin+ client. More services such as a Bluetooth proxy or an AirPlay music proxy could follow later.
• We don't mind whether we use Docker, systemd or the HA Supervisor. The simplest solution that works best wins.
• Existing setups must keep working, so nobody on Docker gets left behind.
• We see PiCompose as a box around another box, and an HA OS box for other devices will follow. Installing add-ons from the HA dashboard via the Supervisor would be nice. We're not sure it's possible, and it isn't a short-term priority.

Hardware (FPH)

1. We're waiting for FPH's feedback before creating a new repo just for hardware control, with a device-based plugin system. FPH wrote a great daemon that we want to use as its foundation. The repo would contain everything except LVA and connect to LVA through the peripheral API.
2. FPH will soon have a .deb that updates the device firmware, and they want to handle it from the HA UI like on ESP32. That needs two buttons, one for LVA and one for firmware, active only on specific hardware. How we detect that is still open.

Audio and voice

• We want to improve voice quality with the technologies available, including beamforming. We're currently looking at the EchoMuse voice pipeline for ideas: https://github.com/wilbowes/EchoMuse/blob/main/docs/voice-pipeline.md
• We'd build all of this as a plugin, so that it can later move out of the device into its own application if OHF needs that.

PipeWire

• We still need to find out whether PipeWire has a real purpose for us or is just a stone on the path to our target. See also Delay in playing the wakeup sound after being idle #346 (delayed wake sound after idle on a Raspberry Pi Zero 2 W).
• If PipeWire's extensions for voice processing and improvement turn out too complicated or not worth it, we'd rather use only the ALSA mixer, which the Sendspin client now uses as well.
• PipeWire could stay available as an optional component for people who want it, especially with regard to the new voice processing that Mike (@synesthesiam) is working on.

Distribution

• We want to set up a PPA repo for the PiCompose configs, so they can be updated.