List end-of-turn operating points

GET/v2/eot/profiles no key needed

Every published operating point, per tier (full is /v2/stream, fast is the table /v2/room uses) and per language, plus the default profile for each conversation mode. A stream session picks one with eot.profile, with eot.latency_budget_ms, or by its mode.

Each point was measured on Baton's end-of-turn benchmark: mean_latency_ms is the mean delay from the end of speech to turn.end, cutoff_rate the share of pauses wrongly ended as turns, cutoff_budget the target that point was set for. Languages without their own table use English.

Responses

200The published operating points.

{
  "tiers": {
    "fast": {
      "en": [
        {
          "name": "cutoff_2pct",
          "mean_latency_ms": 1134,
          "cutoff_rate": 0.0199,
          "cutoff_budget": 0.02
        },
        {
          "name": "cutoff_5pct",
          "mean_latency_ms": 726,
          "cutoff_rate": 0.0482,
          "cutoff_budget": 0.05
        },
        {
          "name": "cutoff_10pct",
          "mean_latency_ms": 439,
          "cutoff_rate": 0.0993,
          "cutoff_budget": 0.1
        }
      ]
    },
    "full": {
      "en": [
        {
          "name": "cutoff_2pct",
          "mean_latency_ms": 875,
          "cutoff_rate": 0.0199,
          "cutoff_budget": 0.02
        },
        {
          "name": "cutoff_5pct",
          "mean_latency_ms": 577,
          "cutoff_rate": 0.0482,
          "cutoff_budget": 0.05
        },
        {
          "name": "cutoff_10pct",
          "mean_latency_ms": 350,
          "cutoff_rate": 0.0936,
          "cutoff_budget": 0.1
        }
      ]
    }
  },
  "mode_defaults": {
    "transactional_call": "cutoff_10pct",
    "working_session": "cutoff_5pct",
    "one_on_one": "cutoff_5pct",
    "brainstorm": "cutoff_2pct",
    "formal": "cutoff_2pct"
  }
}