Skip to content

TTS listening scorecard (JOE-1735)

Not MOS. Use this form for bias-controlled human comparison of Kitten, Kokoro, and future adapters. Record method, listener count, and date.

Session metadata

Field Value
Date
Aurum version / commit
Adapters / models / voices
Listener count
Blinding (yes/no + method)
Hardware

Items (score 1–5)

Fixture id Adapter Intelligibility Naturalness Numbers/abbrev Join artifacts Notes
tts_short
tts_numbers
tts_long_join

Disposition

  • Production-quality candidate? Y/N
  • Blocking issues:
  • Follow-ups:

Objective PCM checks remain separate (aurum_core::eval duration/PCM guards).

Round 001 (recorded)

See repository path evals/reports/listening/listening-round-001.json and the summary in evidence-v004.md. Round 001 is a single-listener, non-blinded pilot — not MOS.