{
  "id": "parakeet-trt-throughput",
  "story": "parakeet-tensorrt-3060",
  "title": "A faster encoder with unchanged transcripts",
  "unit": "times real time",
  "rows": [
    {
      "label": "CUDA FP32",
      "value": 128.8,
      "highlight": false
    },
    {
      "label": "TRT FP16 encoder + FP32 decoder",
      "value": 159.76,
      "highlight": true
    }
  ],
  "scope": "50 clean English clips · 304.515 s · sum of per-clip warm medians",
  "caveat": "Three timing repeats per clip. Aggregate offline speed is not first-word or conversation latency.",
  "kind": "bar",
  "decimals": 2,
  "revision": "7e6fdbe3b1d1aa360346255bfc62945dba694399",
  "sources": [
    {
      "path": "speech-to-text/parakeet-tensorrt-rtx3060-2026-10-05/README.md",
      "url": "https://github.com/groxaxo/experimentos/blob/7e6fdbe3b1d1aa360346255bfc62945dba694399/speech-to-text/parakeet-tensorrt-rtx3060-2026-10-05/README.md",
      "sha256": "ce32eb39e1fa58de14fd32857bb737abf6176f4ec6ce93deb87e1c2c0b9f9e3d"
    }
  ],
  "origin": "reviewed report table",
  "maximum": null,
  "threshold": null
}