ByteSeed Audio 1.0 Launches: One-Liner Prompt Generates Dialogue, Sound Effects, and Ambient Audio

According to WatchNLP, ByteDance has released the Seed Audio 1.0 audio generation model. It can generate dialogue, sound effects, and ambient sound simultaneously based on a single prompt, with timeline-controlled sound cues.

The model supports both text and reference audio inputs with a timing precision of 100ms. It can generate approximately 2 minutes of audio in a single run, extendable while maintaining consistent character voices.

Seed Audio 1.0 supports over 20 languages, allowing voice transfer across languages, as well as changes in tone and emotion.

Official evaluations have shown a audio usability rate of over 90% in most scenarios, with a naturalness MOS score exceeding 4 in the majority of languages. The model is now live in the Volcano Ark Experience Center and offers API access.