# Duplex Dialogue Turn-Taking on Full-Duplex-Bench Source: [Duplex Dialogue Turn-Taking on Full-Duplex-Bench]() Collection index: [llms.txt]() Data updated: 2026-06-09T10:29:52.763Z Comprehensive evaluation of full-duplex conversational capabilities, focusing on Turn-Taking dynamics, Pause Handling, Backchanneling effectiveness, and User Interruption handling, measured by various TOR, Latency, and overall accuracy metrics. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Duplex Dialogue Turn-Taking | | Dataset (including any stated split) | Full-Duplex-Bench | | Metric | Synthetic TOR for Pause Handling | | Unit | SCALAR | | Direction | Lower is better | | Results loaded | 21 of 21 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Synthetic TOR for Pause Handling](): Lower is better; unit: SCALAR. - [Candor TOR for Pause Handling](): Lower is better; unit: SCALAR. - [Backchannel TOR](): Lower is better; unit: SCALAR. - [Backchannel ICC Frequency](): Higher is better; unit: SCALAR. - [Backchannel JSD](): Lower is better; unit: SCALAR. - [Candor TOR for Smooth Turn Taking](): Higher is better; unit: SCALAR. - [Latency for Smooth Turn Taking](): Lower is better; unit: TIME. - [User Interruption TOR](): Higher is better; unit: PERCENT. - [Synthetic GPT-4o Score for User Interruption](): Higher is better; unit: SCALAR. - [Latency for User Interruption](): Lower is better; unit: TIME. - [Averaged Turn-Taking Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Synthetic TOR for Pause Handling | Result date | Report paper | | --- | --- | --- | --- | | GPT-Realtime | 0.01 | Not recorded | 6a279e0890966ade919bf642 | | Listen-Write-Speak | 0.01 | Not recorded | 6a279e0890966ade919bf642 | | DuplexCascade | 0.058 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | Gemini Live | 0.255 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | Gemini Live 2.0 (deprecated) | 0.26 | Not recorded | 6a279e0890966ade919bf642 | | DuplexCascade-β | 0.343 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | PersonaPlex | 0.358 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | PersonaPlex | 0.58 | Not recorded | 6a279e0890966ade919bf642 | | MiniCPM-Duplex | 0.613 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | Freeze-Omni | 0.64 | Not recorded | 6a279e0890966ade919bf642 | | Freeze-Omni | 0.642 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | dGSLM | 0.93 | Not recorded | 6a279e0890966ade919bf642 | | dGSLM | 0.934 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | | Moshi | 0.98 | Not recorded | 6a279e0890966ade919bf642 | | Moshi | 0.985 | Not recorded | 69b0f6fc2d603e6de3e45cf8 | ### GPT-Realtime Synthetic TOR for Pause Handling: 0.01. Result ID: 6a27b03cfbc373592c23a744. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Listen-Write-Speak Synthetic TOR for Pause Handling: 0.01. Result ID: 6a27b03cfbc373592c23a745. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### DuplexCascade Synthetic TOR for Pause Handling: 0.058. Result ID: 69b117876dbb7075b95b11e7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"system backchannels":"false"}. This record does not assert independent reproduction. ### Gemini Live Synthetic TOR for Pause Handling: 0.255. Result ID: 69b117876dbb7075b95b11e6. [Reporting paper]() Recorded evaluation configuration: {"source":"directly obtained from the paper"}. This record does not assert independent reproduction. ### Gemini Live 2.0 (deprecated) Synthetic TOR for Pause Handling: 0.26. Result ID: 6a27b03cfbc373592c23a743. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### DuplexCascade-β Synthetic TOR for Pause Handling: 0.343. Result ID: 69b117876dbb7075b95b11e8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"system backchannels":"true"}. This record does not assert independent reproduction. ### PersonaPlex Synthetic TOR for Pause Handling: 0.358. Result ID: 69b117876dbb7075b95b11e4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"source":"directly obtained from the paper"}. This record does not assert independent reproduction. ### PersonaPlex Synthetic TOR for Pause Handling: 0.58. Result ID: 6a27b03cfbc373592c23a73f. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### MiniCPM-Duplex Synthetic TOR for Pause Handling: 0.613. Result ID: 69b117876dbb7075b95b11e5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"source":"inferred from official checkpoints"}. This record does not assert independent reproduction. ### Freeze-Omni Synthetic TOR for Pause Handling: 0.64. Result ID: 6a27b03cfbc373592c23a742. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Freeze-Omni Synthetic TOR for Pause Handling: 0.642. Result ID: 69b117876dbb7075b95b11e3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"source":"directly obtained from the paper"}. This record does not assert independent reproduction. ### dGSLM Synthetic TOR for Pause Handling: 0.93. Result ID: 6a27b03cfbc373592c23a740. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### dGSLM Synthetic TOR for Pause Handling: 0.934. Result ID: 69b117876dbb7075b95b11e1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"source":"directly obtained from the paper"}. This record does not assert independent reproduction. ### Moshi Synthetic TOR for Pause Handling: 0.98. Result ID: 6a27b03cfbc373592c23a741. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Moshi Synthetic TOR for Pause Handling: 0.985. Result ID: 69b117876dbb7075b95b11e2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"source":"directly obtained from the paper"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.