Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.11329 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911595037196288 |
|---|---|
| author | Züfle, Maike Klejch, Ondrej Sanders, Nicholas Niehues, Jan Birch, Alexandra Lam, Tsz Kin |
| author_facet | Züfle, Maike Klejch, Ondrej Sanders, Nicholas Niehues, Jan Birch, Alexandra Lam, Tsz Kin |
| contents | Spoken conversational systems require more than accurate speech generation to have human-like conversations: to feel natural and engaging, they must produce conversational behaviour that adapts dynamically to the context. Current spoken conversational systems, however, rarely allow such customization, limiting their naturalness and usability. In this work, we present the first open, instruction-following full-duplex conversational speech model that can be trained efficiently under typical academic resource constraints. By keeping the audio encoder frozen and finetuning only the language model, our model requires just 2,000 hours of data, without relying on large-scale pretraining or multi-stage optimization. The model can follow explicit instructions to control speaker voice, conversation topic, conversational behaviour (e.g., backchanneling and interruptions), and dialogue initiation. We propose a single-stage training protocol and systematically analyze design choices. Both the model and training code is released to enable reproducible research on controllable full-duplex speech systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_11329 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | F-Actor: Controllable Conversational Behaviour in Full-Duplex Models Züfle, Maike Klejch, Ondrej Sanders, Nicholas Niehues, Jan Birch, Alexandra Lam, Tsz Kin Computation and Language Spoken conversational systems require more than accurate speech generation to have human-like conversations: to feel natural and engaging, they must produce conversational behaviour that adapts dynamically to the context. Current spoken conversational systems, however, rarely allow such customization, limiting their naturalness and usability. In this work, we present the first open, instruction-following full-duplex conversational speech model that can be trained efficiently under typical academic resource constraints. By keeping the audio encoder frozen and finetuning only the language model, our model requires just 2,000 hours of data, without relying on large-scale pretraining or multi-stage optimization. The model can follow explicit instructions to control speaker voice, conversation topic, conversational behaviour (e.g., backchanneling and interruptions), and dialogue initiation. We propose a single-stage training protocol and systematically analyze design choices. Both the model and training code is released to enable reproducible research on controllable full-duplex speech systems. |
| title | F-Actor: Controllable Conversational Behaviour in Full-Duplex Models |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2601.11329 |