Demo for Scaling Under-Resourced TTS

The audio samples have been systematically organized into five distinct categories, each representing a specific domain:

• Samples 1-5: Randomly selected sentences from Tsync2, encompassing a diverse range of general content.
• Samples 6-10: Sentences pertaining to the legal domain, showcasing content from legal texts.
• Samples 11-15: Sentences drawn from the educational domain, representing academic and instructional content.
• Samples 16-20: Sentences related to the healthcare domain, highlighting content from medical and health-related sources.
• Samples 21-25: Sentences sourced from the financial domain, focusing on economic and financial topics.

This section demonstrates our zero-shot TTS capability.

• Reference Voice: The reference voices that are used as the target for voice cloning.
• Our Model: The output generated by our model when cloning the reference voices.
• OpenVoice: The output generated by the baseline OpenVoice system for comparison.