音源 · Audio credits
This page covers the pronunciation assets in audio/. The exact source for every word is recorded in audio/manifest.json.
Shtooka / University of Caen recordings (104 files)
- Work: Free Audio Collection of Chinese Words (Mandarins)
- Speaker and copyright: Yue Tan, 2009
- Project: Shtooka / SWAC, University of Caen
- Distribution used here:
hugolpz/audio-cmn, commitff9ed3d0c631195bd2c06f39450f3264c7124040 - License: Creative Commons Attribution-ShareAlike 3.0 United States
The selected source MP3s were re-encoded as mono 24 kHz, 64 kbps MP3 and loudness-normalized. Traditional curriculum terms are associated with the corresponding unambiguous simplified-Chinese source filenames in the manifest. These adapted audio files remain available under CC BY-SA 3.0 US. Please keep this attribution, identify further changes, and distribute adaptations of these recordings under the same or a compatible license.
Piper / Chaowen generated pronunciations (48 files)
- Engine: Piper, version 1.7.0, GPL-3.0-or-later (the engine license does not govern its generated audio)
- Voice:
zh_CN-chaowen-medium, pinned at piper-voices commitf5a6e9094787fd865d65cb024472f977f9c542b5 - Voice model repository license: MIT
- Training dataset: OHF-Voice voice-datasets, identified as CC0 in the voice's pinned
MODEL_CARD - Generated locally: 2026-08-19
- Output license: CC0 1.0 Universal
The 48 generated MP3s are released under CC0 1.0 Universal. They replace the previous Microsoft online-service output, which is not included in the public distribution. The selected Piper output was re-encoded as mono 24 kHz, 64 kbps MP3 and loudness-normalized to match the Shtooka assets.
Identifying files
Each manifest entry contains source and license. Shtooka entries also carry the author, pinned source URL, and original source filename; Piper entries carry the engine, voice, model and dataset licenses, pinned model URL, and generation date. This makes mixed-source copies and future replacements auditable without relying on filename guesses.