#Piper custom model

1 messages · Page 1 of 1 (latest)

copper sun
#

I'm trying to use a custom model in wyoming-piper docker container.
I copied both "it_IT-leonardo-high.onnx" and "it_IT-leonardo-high.onnx.json" in my local data folder that is mounted to the data folder inside the container.
I'm running the conteiner with this docker compose file:

wyoming-piper:
    image: rhasspy/wyoming-piper
    network_mode: host
    volumes:
      -  home/satellite/voice/piper/data:/data
    command: >
      --data-dir /data
      --voice it_IT-leonardo-high
    restart: unless-stopped

But it just ignores what I set in the --voice args and downloads the dafault voice for italian, I think because it's the one selected in HA (see screenshot)

#

Also, I modified the it_IT-leonardo-high.onnx.json file to look like this:

{
    "dataset": "it_IT-leonardo-high",
    "audio": {
        "sample_rate": 22050,
        "quality": "High"
    },
    "espeak": {
        "voice": "it"
    },
    "language": {
        "code": "it_IT"
    },
...
formal egret
#

Can't remember offhand but don't think it goes in data, think it's a subdirectory, could check in the container

#

Also you need to pick the voice in ha, that voice argument doesn't apply in HA. The custom voice should be listed in that drop-down if it's detected.

#

You may need to reload the integration for it to appear

formal egret
#

looks like in the container it goes into /config, at least on the container I have. Otherwise, I just have my onnx and onnx.json files in there and they are picked up and show as an option in the HA integration. If you don't see them in the dropdown after reloading the integration, may want to check the logs from the container and see if it is finding the files or not.