VibeVoice 1.5B from Microsoft is made for long-form speech such as podcasts and conversations, where voices need to stay the same from the first line to the last.
Write the script
Put each turn on its own line, starting with Speaker 1: up to Speaker 4:. The tag button inserts these for you. Plain text is spoken by one speaker.
Cast voices
Press Dialogue and choose a voice for each speaker. Cloning needs only audio, no transcript; 10 to 30 seconds works best. Only clone voices you have permission to use.
Quality
A Quality setting trades speed for detail, and Voice likeness sets how closely a take follows the cast voice.
Licence
VibeVoice’s licence and usage notes are set by Microsoft. Read them on the VibeVoice 1.5B model page before you publish anything you make.