README

Awesome-FullDuplexSDM 
A curated list of full-duplex spoken dialogue models.  
 Welcome to PR if you want to add some resources.

 Legend

Type — what the entry actually gives you:

| Type | Meaning |
|:--|:--|
| End-to-end | A standalone full-duplex model: audio in, audio out, duplex behaviour learned inside the model. (omni) means it also takes vision. |
| Cascaded | A complete system assembled from separate parts (streaming ASR + LLM + TTS + a duplex controller). |
| Component | One piece of the puzzle — turn detection, semantic VAD, streaming TTS, retrieval, dialogue management. Needs a host system. |
| Method | A training recipe, reward model, or data scheme rather than a deployable system. |
| Framework | Engineering scaffolding for building/serving duplex systems. |

Open — what has been released:

| Open | Meaning |
|:--|:--|
| Code + weights | Public repo and released checkpoints. |
| Code | Public repo, no checkpoints released (or none linked). |
| API / closed | Usable as a product or API; nothing released. |
| — | Paper or tech report only. |
| ? | Could not verify. |

Year is the year of first public release (arXiv v1, blog post, or repo). Entries without a dated paper are marked with the year they appeared publicly.


 Datasets
| Title | Year | Open | Relevant Resources |
|:--|:-:|:-:|:-:|
| otoSpeech-full-duplex-turn-104h: Full-Duplex Conversational Speech with Turn-Taking Annotations | 2026 | Data | Huggingface/TurnBench/Catalog |
| otoSpeech-full-duplex-task-oriented-20h: Full-Duplex Task-Oriented Conversations | 2026 | Data | Huggingface/Catalog |
| otoSpeech-full-duplex-280h: Full-Duplex Conversational Speech Dataset | 2026 | Data | Huggingface/Catalog |
| MultiTalk: Scaling Full-Duplex Speech Models to Long, Multi-Party, Bilingual Conversation | 2026 | Data | arXiv/Huggingface |
| ConversationalVoice: Full-Duplex Speech Data from Real Conversations | 2026 | Code | arXiv/Github/Demo |
| DuplexGen: Adaptive Synthesis of Human-AI Turn-Taking Dialogues | 2026 | Data + code | arXiv/Github/Huggingface/Demo |
| DuplexChat: Constructing Speaker-Separated Full-Duplex Dialogue Speech at Scale for Spoken Dialogue Language Modeling | 2026 | Data + code | arXiv/Github/Huggingface |
| SOMMELIER: Scalable Open Multi-turn Audio Pre-processing for Full-duplex Speech Language Models | 2026 | Code (pipeline) | arXiv/Demo/Github/Blog |
| SmoothConv & DuplexConv: Large-Scale Chinese Full-Duplex Speech Datasets for Conversational AI | 2026 | Data + code | Github/Demo/Huggingface |
| TURNS-2K | 2025 | Data | Github/Huggingface |



 Models
| Title | Year | Type | Open | Relevant Resources |
|:--|:-:|:--|:-:|:-:|
| Sophon-Realtime v1 | 2026 | Cascaded | Code + weights | Github/HF-Tag/HF-Flash |
| Decode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language Models | 2026 | Method | To be released | arXiv/Github |
| Awaken: Efficient Full-Duplex Adaptation with Capability Retention | 2026 | End-to-end | Code