Step-Audio AQAA
Step-Audio AQAA is a 130B audio-language research release. It generates spoken answers from audio input without a separate transcription-then-synthesis chain.
Research release
Keep two kinds of audio information
The linguistic tokenizer represents phonetic information. The semantic tokenizer preserves acoustic features. The paired streams support direct audio answers and control of spoken expression.StepFun Step-Audio-AQAA model card
Use the release for system research
Practical guidance. This design is useful when studying how an audio model can avoid errors passed between separate ASR and TTS components. It does not remove the need to check the factual content and intelligibility of a spoken answer.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Step-Audio-AQAA |
| License | Apache 2.0.StepFun Step-Audio-AQAA model card |
| Audio representation | Linguistic and semantic token streams are interleaved at a 2:3 ratio.StepFun Step-Audio-AQAA model card |
| Training | The card describes supervised stages, preference optimization and model merging.StepFun Step-Audio-AQAA model card |
Related models
- Step-Audio 2 Mini
StepFun · StepAudio 2
A speech-dialogue model supporting sound understanding, tool calls and retrieval from text and acoustic material.
Launch date not confirmedAPI · Downloadable - Step-Audio Chat
StepFun · Step-Audio
Step-Audio Chat is the 130B multimodal language component of Step-Audio.
Launch date not confirmedDownloadable