ByteDance launches Seed Audio 1.0 for cinema-quality audio

ByteDance's Seed team released Seed Audio 1.0, an audio creation model that combines voice, sound effects, and ambient sounds in one framework to generate cinema-quality audio. The company says sound can now directly tell stories, making listeners 'hear' and 'see' at the same time.

The model allows fine control over multiple elements—one command can create dialogue with emotion, key sound effects, and background sounds, all timed precisely. Voice style is controllable and stays consistent in long audio, with the same voice naturally showing different tones and emotions. It also supports audio generation in over 20 languages, with natural pronunciation and expression per language.

According to official tests, Seed Audio 1.0 shows clear advantages in text-to-voice quality, with over 90% usability in various scenarios. For multilingual audio, most languages scored above 4 on MOS for naturalness, indicating excellent sound quality. Complex instruction-following scored above 3.5 for all languages except Vietnamese. The model is now available on the Volcano Engine experience center.