Create README.md

README.md

@@ -1,3 +1,68 @@
 ---
 license: cc-by-4.0
+language:
+- zh
+- en
+task_categories:
+- automatic-speech-recognition
+tags:
+- code-switching
+- taiwanese-mandarin
+- traditional-chinese
+- mandarin-english
+pretty_name: Taiwan Mandarin–English Code-Switching Sentences
+size_categories:
+- n<1K
 ---
+
+# Taiwan Mandarin–English Code-Switching Sentences (100)
+
+A set of 100 short sentences in **Taiwanese Mandarin with embedded English** (code-switching), spoken by a single Taiwanese speaker. Designed for training and evaluating **Taiwan-specific** Mandarin speech models, with deliberate emphasis on words whose **Taiwan readings differ from Mainland Mandarin**.
+
+## Composition
+
+|             |                                                                                                                               |
+| ----------- | ----------------------------------------------------------------------------------------------------------------------------- |
+| Clips       | 100 (`001.mp3`–`100.mp3`)                                                                                                     |
+| Audio       | MP3, 48 kHz, mono, 256 kbps                                                                                                   |
+| Duration    | 5–14 s each (~13 min total)                                                                                                   |
+| Speaker     | 1                                                                                                                             |
+| `human`     | 20 — recorded by the speaker                                                                                                  |
+| `synthetic` | 80 — generated with [openbmb/VoxCPM2](https://huggingface.co/openbmb/VoxCPM2), using a voice cloned from the human recordings |
+| Language    | Traditional Chinese (zh-Hant, Taiwan) + English                                                                               |
+
+## Format
+
+HuggingFace `AudioFolder` layout: audio files plus `metadata.jsonl`, with one line per clip:
+
+```json
+{"file_name": "048.mp3",
+ "text": "升息的循環還沒結束，房貸的 interest rate 一直在漲，很多人的還款壓力突然變得很大。",
+ "source": "synthetic",
+ "tts_model": "openbmb/VoxCPM2",
+ "reference_audio": "001.mp3",
+ "tts_text": "升{xi2}的循環還沒結束，房{dai4}的 interest rate 一直在{zhang3}，很多人的{huan2}款壓力{tu2}然變得很大。"}
+```
+
+| Field                                        | Meaning                                                                                                                                                                       |
+| -------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `file_name`                                  | Audio file                                                                                                                                                                    |
+| `text`                                       | Clean transcript/label                                                                                                                                                        |
+| `source`                                     | `human` or `synthetic`                                                                                                                                                        |
+| `tts_model` / `reference_audio` / `tts_text` | Provenance for synthetic clips: the model, the speaker’s clip used as the voice reference, and the inline-pinyin text actually fed to the TTS system (`null` for human clips) |
+
+## Pronunciation
+
+The `text` field is the clean transcript/label.
+
+For synthesis, Taiwan readings were enforced with **inline pinyin** (e.g., `企` → `{qi4}`), covering three categories: Taiwan/Mainland pronunciation differences, polyphonic characters, and characters that the TTS system misread and were manually corrected. The synthetic voice’s soft retroflex realization of zh/ch/sh is an intended feature of the Taiwanese accent.
+
+## Intended Use
+
+Training and evaluating **Taiwan Mandarin** speech models, especially on **Mandarin–English code-switching** and Taiwan-specific readings. Use the `source` field to keep human and synthetic data separate, such as evaluating on `human` clips only.
+
+## Limitations
+
+* Single speaker and small dataset.
+* Part of the dataset is synthetic and generated through voice cloning.
+* Synthetic readings were verified through ASR comparison and spot listening, but not exhaustively.