
[EN] A training-ready dataset for prosody (rhythm) pseudonymization research with Style-Bert-VITS2 JP-Extra, built solely from Mozilla Common Voice Japanese (CC0) , release cv-corpus-25.0-2026-03-09 (ja), validated clips only. Contents: 298 speakers / 18,015 utterances (train 14,226 + eval 3,789), 44.1 kHz mono WAV 7-field transcription lists (esd_train/esd_val) with openjtalk phones, tones, and word2ph Per-mora cadence embeddings ({uid}.wav.cadseq.npy, float32 (n_mora,32), L2-normalized, pauses excluded; 11,010 train utterances covered, remainder uses zeros fallback) Training config (config.json) generated from the public JP-Extra template with explicit overrides only (no proprietary-corpus artifacts involved) MANIFEST.sha256 and a format specification (README_bundle.md) enabling users to build their own datasets in the same format Archives unpack directly into the training root (Data/cv_r1/...) of the companion code repository (branch layer-b-cadence-seq). A Colab deployment/verification notebook is provided in the repository. Known limitation: Common Voice "validated" verifies text correctness only, not speaker identity. A listening-based purity audit identified 9 accounts (of 298) containing voices of multiple people. See the datasheet (docs/DATASET_cv_r1.md in the code repository) for details. Related resources: trained model https://huggingface.co/slp-hu/cadence-cv_r1 (AGPL-3.0) · interactive demo https://huggingface.co/spaces/slp-hu/cadence-cv_r1-demo · code https://github.com/slp-hu/Style-Bert-VITS2/tree/layer-b-cadence-seq [JA] 概要cadence cv_r1 は、日本語多話者 TTS およびリズム仮名化研究のための学習データ一式です。Mozilla Common Voice 日本語(cv-corpus-25.0-2026-03-09、validated クリップのみ)のみから構築しており、全体を CC0 で再配布します(298 話者・18,015 クリップ)。 同梱物 cadence_cv_r1_meta_v20260702.tgz(31.9 MiB): 学習 config、ESD リスト(7カラム cleaned 形式)、ケイデンス系列サイドカー 11,010 個(.cadseq.npy、モーラ単位 (n_mora, 32) float32・行 L2 正規化・ポーズ除外)、MANIFEST.sha256、README_bundle.md cadence_cv_r1_wavs_v20260702.tar(4.81 GiB、非圧縮): 音声 18,015 本。アーカイブ内パスは学習ツリー(Data/cv_r1/...)にそのまま一致 再現方法Style-Bert-VITS2 JP-Extra フォーク(slp-hu/Style-Bert-VITS2、ブランチ layer-b-cadence-seq)の Colab ノートブック(deploy/train/eval/synth)から本レコードの URL を直接取得して学習・評価を再現できます。分割は決定的(train は話者あたり ≤200 クリップ、val は評価 split の enroll/trial)。G2P は SBV2 の clean_text 経路と同一です。※同梱の ESD は処理済みのため preprocess_text の再実行は不要です(spk2id が再生成され再現性が壊れます)。 既知の制約CV の validated はテキスト一致の検証であり、話者同一性は保証されません。聴取を含む純度監査で複数人の声を含むアカウント計 9 件を確認しています。詳細な制約・監査結果はフォーク同梱のデータシート(docs/DATASET_cv_r1.md)を参照してください。 ライセンス: データは CC0 1.0(Common Voice ja の原ライセンスに従う)。学習コード(フォーク)および学習済みモデルは AGPL-3.0。 関連リンク: モデル https://huggingface.co/slp-hu/cadence-cv_r1 / デモ https://huggingface.co/spaces/slp-hu/cadence-cv_r1-demo / コード https://github.com/slp-hu/Style-Bert-VITS2/tree/layer-b-cadence-seq
speech synthesis; text-to-speech; voice anonymization; prosody; pseudonymization; Japanese; Common Voice; Style-Bert-VITS2; dataset
speech synthesis; text-to-speech; voice anonymization; prosody; pseudonymization; Japanese; Common Voice; Style-Bert-VITS2; dataset
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
