Spaces:
Running on Zero
Running on Zero
| title: Turn Detection Live Dashboard | |
| emoji: 🎙️ | |
| colorFrom: blue | |
| colorTo: yellow | |
| sdk: gradio | |
| sdk_version: "6.25.0" | |
| python_version: "3.12" | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| # Turn Detection - Live Dashboard | |
| A real-time dashboard for audio turn detection: deciding whether a speaker is | |
| finished talking or just pausing. Speak into the mic or replay a bundled clip | |
| and watch several models score the decision live, plotted against the | |
| waveform. | |
| ## Models | |
| Smart Turn v3.2 (zero-shot), Whisper with trained heads, a Qwen-based semantic | |
| judge, the LiveKit end-of-turn detector, acoustic + semantic fusion, and a | |
| streaming semantic branch. | |
| ## How to use | |
| 1. Select one or more models. | |
| 2. Record from the mic, or replay a sample clip / upload a recording. | |
| 3. Each model's "turn complete" probability is drawn against the waveform, the | |
| dashed line is the decision threshold. | |
| Trained head checkpoints and demo clips are bundled. Other model weights | |
| download on first use. | |