Text-to-Speech API for Piper

[teal epoch]

Is there a text-to-speech API for wyoming-piper similar to that of rhasspy's hermes?

Since I am running HA Supervised (docker addons), I am not sure of the setup for this.
Does this get installed on the host, and runs a separate instance of Piper? Or do I try to install this in the existing addon_core_piper container?

I build a hacky system with netcat:

Inside the addon_core_piper docker container, run a script for a persistent netcat listener that executes piper.
while true do netcat -lvp PORT -c '/usr/share/piper/piper --model /data/en_GB-alan-medium.onnx --output-file -' done

On Homeassistant, I have the "shell_command" integration that runs a script. That script accepts text strings as a variable, and runs echo $1 | nc ADDON_IP PORT > FILE

I have some additional script checks for input validation and stuff. Also, instead of creating a file, I can pipe the output to another netcat to a remote speaker.
I don't need to run HA satellites or media players that need to run anything big. My use case is speaker running from rpi Kodi boxes.

[bold dock]

The Wyoming protocol itself is that API: https://github.com/rhasspy/wyoming
The add-on runs on a TCP port and will follow the TTS message flow: https://github.com/rhasspy/wyoming#text-to-speech-1

[teal epoch]

I was curious about that service listening on 10200?
python3 -m wyoming_piper --piper /usr/share/piper/piper --uri tcp://0.0.0.0:10200

I have seen that accept a type:describe, but not sure if that could also accept tts data.
https://community.home-assistant.io/t/run-whisper-on-external-server/567449

I think I see how it works.
echo '{"type": "synthesize", "data": {"text": "Hello World"} }' | nc ADDON_IP 10200
This does work. Just need to figure out how to pipe the output and close the connection

I think I need the netcat to recognize the audio-stop response payload. I'm guessing that is how the other components do it

-w1 seems to work

Now to figure out how to remove the json data, and keep only the WAV file in a usable format

I guess I would need another tool to convert the raw PCM to WAV format.
I don't see any option in https://github.com/rhasspy/wyoming#text-to-speech-1 to choose format.
Is the Piper HTTP API you linked earlier going to be included in an updated version?

[bold dock]

Sox can convert to WAV. Wyoming doesn't support changing the audio format. That's a client's responsibility.

I hadn't planned on including the HTTP API in the Piper add-on. HA already had a TTS HTTP API: https://www.home-assistant.io/integrations/tts/#rest-api

[teal epoch]

Okay thanks. I'll stick with my hacky netcat workaround. The tts rest api was suggested to me before, but it produces mp3 files which won't work for me. sox is a good idea, but didn't want to add another dependency. Especially since piper on the command line (not api) can already create the wav files that I was used to when using rhasspy.

[bold dock]

It's pretty easy to turn the raw PCM into WAV too. You just need to know the number of samples and the sample rate/width/channels to add the header.

[teal epoch]

The PCM data has the json messages mixed in too... so would need to also remove that then process the PCM into WAV.