Skip to content

Open source CapCut alternative.

TypeScript
5.6 KB
Raw Blame
{
  "cases": [
    {
      "name": "captions/lead-in",
      "what": "The silence before each sentence billed to its first word — the on-device transcriber's habit.",
      "evidence": true,
      "before": {
        "lines": 7,
        "medianStart": 0.004,
        "p90Start": 0.9,
        "worstStart": 0.9,
        "within100ms": 0.571,
        "within250ms": 0.571
      },
      "after": {
        "lines": 7,
        "medianStart": 0.001,
        "p90Start": 0.004,
        "worstStart": 0.004,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.1,
        "p90Start": 0.15,
        "within250ms": 1
      },
      "breaches": []
    },
    {
      "name": "captions/lag",
      "what": "Every cue a quarter second late; the shape is right, the placement is not.",
      "evidence": true,
      "before": {
        "lines": 6,
        "medianStart": 0.251,
        "p90Start": 0.255,
        "worstStart": 0.255,
        "within100ms": 0,
        "within250ms": 0.333
      },
      "after": {
        "lines": 6,
        "medianStart": 0.001,
        "p90Start": 0.005,
        "worstStart": 0.005,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.1,
        "p90Start": 0.15,
        "within250ms": 1
      },
      "breaches": []
    },
    {
      "name": "captions/drift",
      "what": "A clock that slips as the chunk runs on, ending 0.6s out — the hosted transcriber's, unchunked. The tail budget allows one cue: once the slip exceeds the line gap, a caption's true onset sits mid-speech in the previous line's run, which no waveform can testify to. Chunking bounds real drift far under that (cloudTranscribe.ts).",
      "evidence": true,
      "before": {
        "lines": 6,
        "medianStart": 0.305,
        "p90Start": 0.506,
        "worstStart": 0.506,
        "within100ms": 0.167,
        "within250ms": 0.5
      },
      "after": {
        "lines": 6,
        "medianStart": 0.013,
        "p90Start": 0.846,
        "worstStart": 0.846,
        "within100ms": 0.667,
        "within250ms": 0.833
      },
      "budget": {
        "medianStart": 0.12,
        "p90Start": 0.9,
        "within250ms": 0.8
      },
      "breaches": []
    },
    {
      "name": "captions/noisy-room",
      "what": "The same lag over a room 22dB louder — a noise floor close enough to the voice to blur its edges.",
      "evidence": true,
      "before": {
        "lines": 6,
        "medianStart": 0.251,
        "p90Start": 0.255,
        "worstStart": 0.255,
        "within100ms": 0,
        "within250ms": 0.333
      },
      "after": {
        "lines": 6,
        "medianStart": 0.002,
        "p90Start": 0.01,
        "worstStart": 0.01,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.12,
        "p90Start": 0.2,
        "within250ms": 1
      },
      "breaches": []
    },
    {
      "name": "captions/music-bed",
      "what": "A steady bed under the whole cut: the gaps between lines are the only readable quiet.",
      "evidence": true,
      "before": {
        "lines": 6,
        "medianStart": 0.251,
        "p90Start": 0.255,
        "worstStart": 0.255,
        "within100ms": 0,
        "within250ms": 0.333
      },
      "after": {
        "lines": 6,
        "medianStart": 0.001,
        "p90Start": 0.005,
        "worstStart": 0.005,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.15,
        "p90Start": 0.3,
        "within250ms": 0.9
      },
      "breaches": []
    },
    {
      "name": "captions/fast-cadence",
      "what": "Lines a quarter second apart, where a snap that overshoots lands on the wrong line.",
      "evidence": true,
      "before": {
        "lines": 6,
        "medianStart": 0.005,
        "p90Start": 0.4,
        "worstStart": 0.4,
        "within100ms": 0.667,
        "within250ms": 0.667
      },
      "after": {
        "lines": 6,
        "medianStart": 0.003,
        "p90Start": 0.005,
        "worstStart": 0.005,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.12,
        "p90Start": 0.2,
        "within250ms": 1
      },
      "breaches": []
    },
    {
      "name": "lyrics/half-heard",
      "what": "The user's lyrics against a transcript that missed a third of the words and misheard a quarter.",
      "evidence": true,
      "before": {
        "lines": 8,
        "medianStart": 0.193,
        "p90Start": 0.697,
        "worstStart": 0.697,
        "within100ms": 0,
        "within250ms": 0.625
      },
      "after": {
        "lines": 8,
        "medianStart": 0.004,
        "p90Start": 0.006,
        "worstStart": 0.006,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.12,
        "p90Start": 0.25,
        "within250ms": 0.9
      },
      "breaches": []
    },
    {
      "name": "lyrics/mostly-unheard",
      "what": "Singing the recognizer barely follows: two words in three unheard, so most line times are interpolated.",
      "evidence": true,
      "before": {
        "lines": 8,
        "medianStart": 0.296,
        "p90Start": 0.84,
        "worstStart": 0.84,
        "within100ms": 0.125,
        "within250ms": 0.5
      },
      "after": {
        "lines": 8,
        "medianStart": 0.004,
        "p90Start": 0.006,
        "worstStart": 0.006,
        "within100ms": 1,
        "within250ms": 1
      },
      "budget": {
        "medianStart": 0.15,
        "p90Start": 0.35,
        "within250ms": 0.85
      },
      "breaches": []
    }
  ],
  "failed": []
}