{
  "note": "Golden STT fixtures. Audio is piper-synthesised, not recorded — see scripts/gen-stt-fixtures.sh, which reads `text` from this file and synthesises from it. This is the only source of the spoken words; regenerate with that script and do not hand-edit `wav`.",
  "wer_note": "max_wer is set just above what each case actually measures against ggml-small, recorded in `measured_wer` on 2026-08-01. A flat 0.34 over a five-word reference tolerated two wrong words and left most of the range unguarded. A model swap should show up as a diff to these numbers, not as silence: rerun `make test-stt-golden`, read the logged transcript, and move both fields together.",
  "cases": [
    {
      "name": "ru_reminder",
      "wav": "ru_reminder.wav",
      "lang": "ru",
      "text": "напомни мне через час позвонить маме",
      "keywords": ["напомни", "час", "позвонить"],
      "measured_wer": 0.0,
      "max_wer": 0.1
    },
    {
      "name": "ru_fact",
      "wav": "ru_fact.wav",
      "lang": "ru",
      "text": "отметь что я выпил воды",
      "keywords": ["отметь", "воды"],
      "measured_wer": 0.2,
      "max_wer": 0.25
    },
    {
      "name": "ru_query",
      "wav": "ru_query.wav",
      "lang": "ru",
      "text": "что у меня сегодня по календарю",
      "keywords": ["сегодня", "календарю"],
      "measured_wer": 0.0,
      "max_wer": 0.1
    },
    {
      "name": "en_act",
      "wav": "en_act.wav",
      "lang": "en",
      "text": "restart the web server and check the disk space",
      "keywords": ["restart", "server", "disk"],
      "measured_wer": 0.0,
      "max_wer": 0.1
    }
  ]
}