Today ElevenLabs released their new Text-to-speech (TTS) model. Its called v3 and even though its still in beta mode the results are insane!
Until this point configurations and customisations of TTS or Realtime models for small languages are still quite poor.
I come from Slovenia and with my team at brainylab we try to build custom Voice AI tools (or agents) for quite some time now.
We had some success, but its quite tedious work configuring voice models to speak somewhat good Slovenian language.
When I saw the Elevenlabs annuncement I tested it out to see how much improvemenet was done there and let me tell you, this is getting crazy!
This is first experiment with completely default values.
If you’re not slovenian speaking person, this is the text the model is conversating:
Voice 1: Turn every second website visitor into active participant. Slovenian tool, knowledge and technology that turns website visitors in your communication and makes them active amabasdors of your brand.
Voice 2: With the use of gamification you will include over 50% ob website visitors into active participants.
But not only the Slovenian pronunciation and tonality are getting amazing, you can also annotate the model to give it even more expressions. (text is the same as with the first recording)
Sure, there are some bugs still, but the model is in beta and still works like this.
