Some voices refuse to install with invalid archive error

[pitermach]

To reproduce:

• Download the Polish voice - https://github.com/rhasspy/piper/releases/download/v0.0.2/voice-pl-mls_6892-low.tar.gz
• Attempt to install it from the Piper settings After selecting the file, I get an installation failed message and the log shows the following:

ERROR - external:globalPlugins.piper_voice_installer.SettingsPanel._get_process_tar_archive (12:02:32.001) - MainThread (8388):
Failed to install voice from archive
Traceback (most recent call last):
  File "C:\Users\piter\AppData\Roaming\nvda\addons\piper_neural_voices\globalPlugins\piper_voice_installer.py", line 78, in _get_process_tar_archive
    voice_key = PiperTextToSpeechSystem.install_voice(filepath, PIPER_VOICES_DIR)
  File "C:\Users\piter\AppData\Roaming\nvda\addons\piper_neural_voices\synthDrivers\piper_neural_voices\tts_system.py", line 280, in install_voice
    raise ValueError(f"Invalid voice archive: `{archive_path}`")
ValueError: Invalid voice archive: `D:\Downloads\voice-pl-mls_6892-low.tar.gz`

The Polish voice isn't the only one which fails in this way - I got an identical error when trying to install the Southern english voice. Others, like US English Lessac, Ryan and Danny, install with no problems.

[mush42]

@pitermach

For now, just rename the voice file to have the following form: voice-language-name-quality

replace language with the voice's language (e.g en-GB), name with the voice's name (e.g amy), and quality with the voice's quality (e.g low).

I'll fix this soon.

Best Musharraf

[pitermach]

Thanks, I was able to get the voices installed following your instructions.

[CyrilleB79]

Same issue with "voice-fr-mls_1840-low.tar.gz".

Renaming the voice allows to install it. But there are still errors, e.g. opening the variant combobox.

[KevanGP]

En-GB Southern English Female will not install. I m not sure where the logs are located or I'd paste the error. I did try redownloading the voice with same results.

[mush42]

@KevanGP please take a look at my first comment above.

[tmthywynn8]

Would x-low voices populating improper fields be related to this issue? For example, installing the Vietnamese vivos voice dataset which does have a model card) produces the following:

Name	Quality	Language
X	Low	vi-VIVOS

Voice model card button output:

[Window Title]
Info

[Content]
Model card information is unavailable for this voice

[OK]