Is there a way or are there plans to introduce emotions?

[juangea]

Hi there.

I'm using Melo and the quality is very good, in spanish works very well, however it's emotion-less, it feels dead, it's greaet for some uses, but for others is very monotone, is there a way to introduce amotion or are there any plans to do so?

Thanks!

[dezynetechnologies]

I guess you can use tone-coloring from OpenVoice..We have used meloTTS for creating AI generated voices using clear recording of a single speaker for around 40-45 minutes. We trained for around 1000 epochs.

[juangea]

How could we use tone coloring?

The reference audio changes the tone and the emotion? I thought it was only trying to clone the voice, but not the emotion.

For us, Melo is giving good results, in relation to clarity of speech, the problem is that the entonation is too emotionless and monotone, and that's what we are trying to solve.

The trained voice you trained, sound more natural?

[SayanoAI]

All open source TTS models will be emotionless because the data they are trained on are emotionless (e.g. librispeech). Close sourced models (like openai and elevenlabs) can get away with using copyrighted data or paying people for emotional voice data to train their models.

[juangea]

I don't agree with that asumption.

There are techniques to introduce emotion, like the one used by F5, also Bark has emotion on it, not sure their data set but it's open source.

Emotion can be introduced by providing a reference and then translating that reference over to the inference, the data set it's not so important with this techniques.

Toucan does something similar too.