声译 Shengyi — Data sources

Audio, artwork and timestamped lyrics are retrieved from NetEase Cloud Music public services.
Only publicly available full playback URLs for the catalog versions are resolved.
There is no account-cookie forwarding or paywall unlocking. The app does not host audio files.
Actual performers and versions are shown individually.
Audio availability is controlled by the provider and may change by region or time.

Factual word readings: 教育部臺灣閩南語常用詞辭典, via g0v/moedict-data-twblg.
https://github.com/g0v/moedict-data-twblg
Dictionary content: CC BY-ND 3.0 TW. Formatting contribution: Audrey Tang, CC0.
Only the title and unchanged first listed reading are retained in readings.json.
The Mandarin character cues are generated separately by Shengyi and remain approximate.
Context-sensitive phrase rules supplement the unchanged source readings. A separate presentation layer now maps dialect syllables to readable, toneless Mandarin
sound approximations using Mandarin characters, with Mandarin pinyin where needed;
it does not read the original lyric characters in Mandarin.
The underlying Tai-lo/Jyutping is retained as an optional advanced display. There is no recording-specific
listening audit or automatic singing/sandhi transcription. Dictionary definitions are not copied.

Simplified-to-traditional character data: OpenCC, BYVoid and contributors.
https://github.com/BYVoid/OpenCC
Apache License, Version 2.0: https://www.apache.org/licenses/LICENSE-2.0
The STCharacters text table was reformatted as JSON, selecting its first listed variant.

Typography, if available: Noto Serif SC via Google Fonts, SIL Open Font License.
System serif fonts are used when the font service is unavailable.

Cantonese readings: CanCLID / Rime Cantonese, CC BY 4.0.
See CANTONESE-NOTICE.txt for source version, attribution and selection changes.
https://github.com/rime/rime-cantonese
https://creativecommons.org/licenses/by/4.0/

2026-10-21 pronunciation update: hokkien-overrides.json contains factual reading selections
and spelling aliases based on MOE headwords and examples, not copied definitions.
cantonese-overrides.json selects additional CanCLID phrase readings and common defaults,
with manual polyphone corrections; its source and modification attribution is included
in CANTONESE-NOTICE.txt. See READING-NOTES.txt for the text-only review scope.

Mandarin approximation tables (mandarin-hokkien.json and mandarin-cantonese.json)
are editorial sound approximations made for this app, not an official romanization scheme.
They do not replace source dictionary readings or assert recording-level phonetic accuracy.