Meta publishes alignment-free Text-Audiobox research
Meta AI has posted a paper on an alignment-free Text-Audiobox for voice dubbing and full-duplex dialogue synthesis; only the title is available so far.
A new entry has appeared on Meta's AI research site: Alignment-Free Text-Audiobox for Voice Dubbing and Full-Duplex Dialogue Synthesis. The title points to two tasks: voice dubbing and full-duplex dialogue synthesis, where both sides of a conversation can speak at the same time. "Alignment-free" indicates the approach does not rely on pre-computed alignment between text and audio.
Beyond the title, the source we have carries no technical detail. Model size, training data, comparison results, availability and publication date are not stated, so we make no claims about the system's performance here.
The full paper is available through the source link.
