Can not load tts_models/multilingual TTS models

[edisonzf2020]

LocalAI version: 2.15.0

Docker: image: quay.io/go-skynet/local-ai:master-cublas-cuda12-ffmpeg
Environment, CPU architecture, OS, and Version:

Ubuntu 22.04 GPU
Describe the bug

test 3 models :
1: tts_models/multilingual/multi-dataset/xtts_v2
2: tts_models/multilingual/multi-dataset/xtts_v1.1
4: tts_models/multilingual/multi-dataset/bark

curl -L http://192.168.1.19:8090/tts \
    -H "Content-Type: application/json" \
    -d '{
"backend": "coqui",
"input": "Bonjour Madame ! Comment allez-vous ?",
"model": "tts_models/multilingual/multi-dataset/xtts_v2",
"voice": "Ana Florence"
}' | aplay -D pipewire -

To Reproduce

8:25AM ERR Server error error="could not load model (no success): Unexpected err=UnpicklingError(\"invalid load key, '<'.\"), type(err)=<class '_pickle.UnpicklingError'>" ip=172.22.0.1 latency=13.523714559s method=POST status=500 url=/v1/audio/speech
8:25AM INF Loading model 'tts_models/multilingual/multi-dataset/bark' with backend coqui
8:26AM INF Loading model 'tts_models/multilingual/multi-dataset/xtts_v2' with backend coqui
8:26AM ERR Server error error="could not load model (no success): Unexpected err=UnpicklingError(\"invalid load key, '<'.\"), type(err)=<class '_pickle.UnpicklingError'>" ip=172.22.0.1 latency=13.108962696s method=POST status=500 url=/v1/audio/speech
8:26AM INF Loading model 'tts_models/multilingual/multi-dataset/bark' with backend coqui
8:26AM ERR Server error error="could not load model (no success): Unexpected err=RuntimeError('input(): lost sys.stdin'), type(err)=<class 'RuntimeError'>" ip=192.168.1.28 latency=6.335075199s method=POST status=500 url=/tts

Expected behavior

Logs

Additional context

[blob42]

Hi. This will be solved when #2308 is merged.

[bdqfork]

Hi. This will be solved when #2308 is merged.

However, I am still encountering this bug in the latest version 2.19.4.

[quachc]

edit
I checked Discord and it appears you have to accept the terms & conditions for coqui, which you can do with an environment variable; COQUI_TOS_AGREED=1.
Using docker run, add --env COQUI_TOS_AGREED=1

same issue for me with 2.23.0 (and using the exact request as shown in this bug).

DBG GRPC(tts_models/multilingual/multi-dataset/xtts_v2-127.0.0.1:41529): stderr Preparing models, please wait
ERR Server error error="failed to load model with internal loader: could not load model (no success): Unexpected err=RuntimeError('input(): lost sys.stdin'), type(err)=<class 'RuntimeError'>"