Cantonese Voice Benchmark粵語語音基準

Record

What changed, and when.

A benchmark that publishes numbers needs somewhere to publish the sentence explaining why one of them moved. Every correction, every change to how we measure, and every result lands here with a date on it.

ResultsA number has been published or has moved, with the reason.
MethodologyHow we measure has changed. These are versioned and never silent.
AnnouncementSomething is happening, or has changed about the programme.
NoteWorking detail worth writing down.
  1. The pool now accepts audio in the formats phone systems actually produce

    The first real recording handed to us was GSM 6.10, which no browser will decode.

  2. Withdrawing is now something you do yourself

    The rules promised it. Until now it required emailing someone, which is a different promise.

  3. A system description is now required before final scoring

    One to two pages on what was measured. The Full Score is withheld until it is in.

  4. Entries whose intervals overlap are published as tied

    With thirty clips a fraction of a percent is noise, and ordering noise is how a leaderboard becomes theatre.

  5. The text normalization is published as running code

    Prose describing how a transcript is cleaned before scoring is not enough to reproduce a number.

  6. Why this benchmark exists, and what it is not

    Cantonese speech evaluation is not missing. Evaluation on real Hong Kong enterprise call audio is.