Built-in support to download TTS models en masse?

[AndrewBarfield]

I'm creating a Docker container for TTS and would like to be able to download all models/vocoders in one fell swoop.

I've found: https://github.com/coqui-ai/TTS/blob/main/TTS/.models.json, so I could write a script to do it but that could get stale quickly.

Is there a command-line argument to do this for TTS or TTS-Server?

Thanks in advance!

(Completely unrelated P.S.: Is the default voice Korva Coleman?)

[erogol]

Default is LJSpeech dataset.

There is no command to download all the models. You need to call once each model or you can use ModelManager class to do that manually.

[meonkeys]

@AndrewBarfield would you mind sharing your Dockerfile/project? I'm trying to modify and build https://github.com/synesthesiam/coqui-docker/ for amd64 (CPU) only but I haven't got it working yet.

[AndrewBarfield]

Hope this helps!

Dockerfile.txt

[meonkeys]

thank you! Sure does. Is fix-permissions anything special? Maye just a recursive chown? I guess it is a separate script.

[AndrewBarfield]

It ensures everything is accessible via the coqui user. Here it is:
fix-permissions.txt

This general dockerfile setup/architecture (including 'fix-permissions') was taken from https://github.com/jupyter/docker-stacks.

[AndrewBarfield]

Oh, BTW, here's the script that I use to download all the models.

#!/bin/bash

list=$(tts --list_models | grep -oP 'tts_models[^\s]*')

while IFS= read -r line; do
	filename=$(basename -- "$line")
	tts --text "Hello world!" \
		--model_name "$line" \
		--out_path "wav/$filename.wav"
	echo -e "\n"
done <<< "$list"

[meonkeys]

Thanks again!

[meonkeys]

Have you seen https://www.web3.lu/text-to-speech-sound-samples-from-coqui-tts/ ? I found that in #522

[brandonscript]

Updated script, this one filters by language (change it from /en to whatever you want, or remove that bit from lines 3 and 11) and handles --list_models returning values that start with ##:.

#!/bin/bash

list=$(tts --list_models | grep "tts_models/en/")

out_path="./path/to/output"

mkdir -p "$out_path"

while IFS= read -r line; do
        model=$(echo "$line" | sed -E 's/^ [0-9]+: //' | sed -E 's/ \[already downloaded\]$//')
	filename=$(echo "$model" | sed -E 's/^tts_models\/en\///' | sed -E 's/\//_/g')
	cmd="tts --text \"Hello world!\" \
--model_name \"$model\" \
--out_path \"$out_path/$filename.wav\""
	eval "$cmd"
done <<< "$list"

[ErcinDedeoglu]

You must agree to the terms of service to use this model.
| > Please see the terms of service at https://coqui.ai/cpml.txt
| > "I have read, understood and agreed the Terms and Conditions." - [y/n]
| | > Traceback (most recent call last):
File "/usr/local/bin/tts", line 8, in
sys.exit(main())
File "/root/TTS/bin/synthesize.py", line 424, in main
model_path, config_path, model_item = manager.download_model(args.model_name)
File "/root/TTS/utils/manage.py", line 403, in download_model
self.create_dir_and_download_model(model_name, model_item, output_path)
File "/root/TTS/utils/manage.py", line 331, in create_dir_and_download_model
raise Exception(" [!] You must agree to the terms of service to use this model.")
Exception: [!] You must agree to the terms of service to use this model.