mtsmfm.github.io

correct_segments

Fix garbled speech recognition in segments that were already transcribed. Give the segment's t0 (as you received it, in seconds) and the text as it should read; the page keeps the raw ASR output in text and stores yours in corrected, shows corrected in the 文字起こし / Transcript panel with a 修正 / fixed badge (the original stays visible as a tooltip) and saves it to transcript.jsonl. An empty text clears a previous correction. Use it for proper nouns and mis-heard words once the context makes the right reading obvious — and call add_glossary too, so later utterances come out right in the first place. t0 is matched after rounding to 0.1 s; t0 values that match no segment come back in unmatched. Returns {ok, updated, unmatched}.

Page: /ramble-to-slides/

Input schema

{
  "type": "object",
  "properties": {
    "corrections": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "t0": {
            "type": "number",
            "description": "The segment's t0, in seconds from recording start."
          },
          "text": {
            "type": "string",
            "description": "The corrected text. Empty clears the correction."
          }
        },
        "required": [
          "t0",
          "text"
        ]
      },
      "description": "One entry per segment to correct."
    }
  },
  "required": [
    "corrections"
  ]
}