Shinoda Suzuki K.

Script & coverage: phoneme-balanced prompts (JP kana/kanji, EN, PT-BR), numbers/dates, loanwords, code-switching. Direction & recording: mic technique, pace/energy consistency; delivery in WAV 48kHz/24-bit, noise-profiled. Annotation: timestamps, disfluencies/breaths, IPA notes, prosody tags/SSML, emotion labels. Data QA: SNR/clipping checks, loudness targets (e.g., −23 or −16 LUFS), silence trimming, alignment/segmentation. Evaluation: MOS test design, ABX, WER/CER test sets incl. edge cases and domain terms. Tooling: Praat, Audacity/Adobe Audition, Label Studio/ELAN, ffmpeg, Whisper, Python (pandas/regex), spreadsheets. Compliance: consent management, PII redaction, licensing for training data. Also: prompt packs for AI voice creators; basic post-processing; light video sync when needed. Deliverables Curated audio + CSV/JSON labels SSML-ready scripts QA report with metrics (SNR, loudness, error rates) Eval sets + scoreboard template