ByteDance says Seed Audio 1.0 is available through BytePlus

The company says its audio model generates speech, sound effects and ambience together from a prompt, with controls for dialogue timing.
In this article
Sole source: ByteDance says its Seed Audio 1.0 audio creation model is available through BytePlus. In its November 16 announcement, the company said a single prompt can direct the model to produce speech, sound effects and ambience together, coordinating performance, timing and background sound.
ByteDance says much audio production still involves making separate layers, placing them on a timeline and repeatedly adjusting their timing and mix. Its model instead aims to generate those elements as a single scene, an approach the company says could serve dubbing, advertising, games and podcasts.
Voices, languages and timing
The company says the model can generate up to 2 minutes of audio in one pass and continue beyond that. Creators can shape a voice using a text description, an authorized reference sample or both, without training a separate model for each speaker.
ByteDance says the model supports generation in more than 20 languages and can retain a character voice’s recognizable sound and performance style across them. It says creators can specify when dialogue begins at 100-millisecond intervals, a feature the company says could serve video dubbing, re-voicing, advertising and scripted audio where timing matters. Fine timing control currently focuses mainly on dialogue.
The company’s evaluations
In its own subjective comparisons, ByteDance says the model’s text-described voices gained preference and its usability and high-quality output rates improved. The company says usable audio exceeded 90% in most scenarios it evaluated. These are company evaluations, not independent tests.
In its human evaluation of multilingual audio, ByteDance says most languages scored above 4.0 for naturalness on a mean opinion score, an average of human ratings. For following instructions in complex audio tasks, every evaluated language except Vietnamese scored above 3.5.
Planned controls
ByteDance plans to extend fine control to sound effects, ambience and music, and to improve voice consistency in longer, more complex scenes. It also plans to support video references and explore long-form and multitrack audio generation, as well as multilingual translation that lets creators control expression, timing and duration. The company has not given delivery dates.
Follow AveNews
Earlier in Technology
SuperintelligenceByteDance launches Seedance 2.5 video model
ByteDance says the model is rolling out on Jimeng AI and Doubao Pro and can generate up to 30 seconds of video at a time, up from 15 seconds.
2 min
SoftwareApple launches health question study with OpenEvidence
Eligible US adults aged 21 or older can enroll with an Apple Intelligence-capable iPhone running iOS 26 or later. They can ask health questions, read answers with clinical citations and rate them. Apple says the app is not intended to diagnose or treat conditions.
2 min
CompaniesSK Hynix plans 1.4 trillion won (about $1 billion) supplier support
The chipmaker plans to support South Korean suppliers over five years and says a jointly funded testing facility, Trinity Fab, is scheduled to open in May 2027.
3 min
