Help us choose a default voice!

[JarbasAl]

ovos-core can run fully offline and without backend, you might not even have a pairing step

This means we need a default value for voice that works offline!

We also need to have a fallback_tts value, this one is used when the main TTS fails.

It's probably ok if we use the same TTS since the users will likely only change the main voice and this is the voice we always ensure is available

We will always need some binary installed, none of the local tts plugins are pure python, as long as it is FOSS and easily available we can consider any voice

At some point we might want to have a default per language, that will require code changes but we can start collecting data to inform that decision

Add more voices to the list below!

• Alan Pope - Mimic - "en-US" - Male
• PicoTTS - "de-DE", "es-ES", "fr-FR", "it-IT", "en-US" - Female
• m1 - espeak-ng - all languages - Male
• f1 - espeak-ng - all languages - Female
• ...

[builderjer]

I like the voices from Larynx.
https://github.com/NeonGeckoCom/neon-tts-plugin-larynx_server/tree/refactor/bump_opm

Especially if this works.

[ChanceNCounter]

https://rhasspy.github.io/larynx/#en-us

• ljspeech
• southern_english_female
• southern_english_male

all had, I thought, compelling samples.

[Akul2010]

The espeak male voices sound better than the female voices. They both are kind of robotic, though.

[kha84]

Hey! Sorry, I'm late to the party :)

I just recently stumbled upon OpenVoiceOS and I really love the direction you're guys moving it into ❤️

As for the standard voice, some time ago, I was playing with different TTS models using Coqui API and I've made this simple script to run-them-all and generate a report to compare them. I was primarily looking for the ones, which are giving the best audio results, while keeping the least pressure on hardware. Have a look - https://github.com/kha84/tts-comparison

[JarbasAl]

I am adding existing plugins to this repo https://github.com/OpenVoiceOS/ovos-tts-auto-cache

all samples there are relevant

[andlo]

I like Allan Pope voice, as it sounds like Marvin the Paranoid Android from The Hitchhiker's Guide to the Galaxy :)

[erikjber]

I personally prefer en_US/hifi-tts_low speaker 92. She sounds British and therefore classy.

[builderjer]

I use `en_US/vctk_low#s5 because of the same reason...and my wife likes it. I will check that one also