Update README with new dataset card

README.md

@@ -3,65 +3,208 @@ language:
   - en
   - zh
 license: cc-by-nc-4.0
-task_categories:
-  - translation
-  - automatic-speech-recognition
-pretty_name: MoVE — Mixture of Vocalization Experts Dataset
+pretty_name: MoVE — Multilingual Voice Emotion (paired EN/ZH)
 size_categories:
   - 100K<n<1M
+task_categories:
+  - audio-to-audio
+  - text-to-speech
+  - translation
 tags:
-  - speech-to-speech-translation
-  - expressive-speech
-  - emotion
-  - bilingual
-  - tts
+  - speech-translation
+  - emotional-tts
+  - paired-speech
+configs:
+  - config_name: "1hr"
+    data_files:
+      - split: train
+        path: metadata_1hr.tsv
+  - config_name: "50hr"
+    data_files:
+      - split: train
+        path: metadata_50hr.tsv
+  - config_name: "100hr"
+    data_files:
+      - split: train
+        path: metadata_100hr.tsv
+  - config_name: "500hr"
+    data_files:
+      - split: train
+        path: metadata_500hr.tsv
+  - config_name: "1000hr"
+    data_files:
+      - split: train
+        path: metadata_1000hr.tsv
 ---
 
-# MoVE Dataset
+# MoVE — Multilingual Voice Emotion (paired EN/ZH)
+
+Repo: [`47z/MoVE`](https://huggingface.co/datasets/47z/MoVE)
+
+Paired English / Chinese emotional speech for speech-to-speech translation
+training. Every row is a `(en_wav, zh_wav)` pair belonging to one of five
+emotion categories: **angry, crying, happy, laugh, sad**.
+
+## Subsets (nested, emotion-balanced)
+
+| subset | en hours | pairs   | per-emotion hours | balanced? |
+|-------:|---------:|--------:|------------------:|:---------:|
+| 1hr    | 1.00     | 839     | 0.20 each         | ✓ |
+| 50hr   | 50.00    | 42,467  | 10.00 each        | ✓ |
+| 100hr  | 99.99    | 85,004  | 20.00 each        | ✓ |
+| 500hr  | 500.00   | 425,952 | 100.00 each       | ✓ |
+| 1000hr | 1055.67  | 896,477 | 95–133            | ✗ (full data) |
+
+Smaller subsets are strict subsets of larger ones. Hours are measured by
+**English** audio duration; the Chinese side is paired one-to-one and
+~5% longer in total.
 
-Bilingual (EN↔ZH) expressive speech-to-speech translation dataset.  
-858,312 pairs · 5 emotion categories: `angry`, `happy`, `sad`, `laugh`, `crying`.
+## Repository layout
 
-> **Paper:** [MoVE: Translating Laughter and Tears via Mixture of Vocalization Experts in Speech-to-Speech Translation](https://arxiv.org/abs/2604.17435) (Interspeech 2026)
+```
+move-1000hr/
+├── README.md
+├── metadata_1hr.tsv       # 5 columns: zh_path, en_path, zh_text, en_text, category
+├── metadata_50hr.tsv
+├── metadata_100hr.tsv
+├── metadata_500hr.tsv
+├── metadata_1000hr.tsv    # full
+├── metadata_all.tsv       # all 896k pairs + 5 boolean cols (in_1hr ... in_1000hr)
+├── plan.tsv               # full plan with tier + shard_id + durations
+├── entries.jsonl.gz       # lossless original metadata (optional)
+├── data/                  # audio packed into tar shards
+│   ├── s1hr_000.tar
+│   ├── s50hr_*.tar
+│   ├── s100hr_*.tar
+│   ├── s500hr_*.tar
+│   └── s1000hr_*.tar
+└── scripts/
+    └── tsv_to_metadata_shard.py   # paired TSV -> conversation-format JSONL
+```
 
-## Directory Structure
+Inside each tar, paths preserve the original tree:
 
 ```
-├── en/{emotion}/*.flac       # English TTS audio (FLAC, lossless)
-├── zh/{emotion}/*.flac       # Chinese TTS audio (FLAC, lossless)
-├── metadata.tsv              # Pair metadata (see below)
-└── make_kimi_train.py        # Convert to Kimi-Audio training format
+en/<emotion>/<id>_en.wav
+zh/<emotion>/<id>_zh.wav
 ```
 
-## metadata.tsv
+So `tar -xf data/*.tar` reproduces a flat `en/` + `zh/` directory layout.
 
-Columns: `zh_path`, `en_path`, `zh_text`, `en_text`, `category`
+## Quick start
 
-All paths are relative to this directory and use `.flac` extension.
+### Download just one subset (e.g. 100hr ≈ 34 GB audio)
 
 ```bash
-python make_metadata_tsv.py
+huggingface-cli download 47z/MoVE \
+  --repo-type dataset --local-dir ./move \
+  --include "metadata_100hr.tsv" \
+            "data/s1hr_*.tar" \
+            "data/s50hr_*.tar" \
+            "data/s100hr_*.tar" \
+            "scripts/*"
+
+cd move
+for f in data/*.tar; do tar -xf "$f"; done   # extracts en/ and zh/
 ```
 
-## Kimi-Audio Training Format
+The wav paths in `metadata_100hr.tsv` (e.g. `en/happy/happy_000001_en.wav`)
+now resolve to local files.
+
+### Download the full 1000hr (~340 GB)
+
+```bash
+huggingface-cli download 47z/MoVE \
+  --repo-type dataset --local-dir ./move
+cd move && for f in data/*.tar; do tar -xf "$f"; done
+```
+
+### Required tar shards per subset
+
+| subset  | tar prefixes to download |
+|---------|--------------------------|
+| 1hr     | `s1hr_*` |
+| 50hr    | `s1hr_*`, `s50hr_*` |
+| 100hr   | `s1hr_*`, `s50hr_*`, `s100hr_*` |
+| 500hr   | `s1hr_*`, `s50hr_*`, `s100hr_*`, `s500hr_*` |
+| 1000hr  | all (`s*_*`) |
 
-> ⚠️ **WARNING**: `make_kimi_train.py` converts all `.flac` files to `.wav` **in-place**
-> and deletes the original `.flac` files. WAV files are approximately **2–3× larger**
-> than FLAC. Run this only when you intend to use the data for local Kimi-Audio training
-> and no longer need the FLAC files.
+## metadata.tsv schema
 
-Produces `metadata_kimi_train.jsonl` in Kimi-Audio conversation format:
+Each `metadata_<size>.tsv` is tab-separated with this header:
 
-```json
-{"task_type": "s-s", "conversation": [
-  {"role": "user", "message_type": "text", "content": "Translate the given English speech into Chinese while preserving its expressiveness."},
-  {"role": "user", "message_type": "audio", "content": "en/angry/angry_000001_en.wav"},
-  {"role": "assistant", "message_type": "audio-text", "content": ["zh/angry/angry_000001_zh.wav", "..."]}
-]}
+| column   | example | description |
+|----------|---------|-------------|
+| zh_path  | `zh/happy/happy_000001_zh.wav` | path to Chinese wav (relative) |
+| en_path  | `en/happy/happy_000001_en.wav` | path to English wav (relative) |
+| zh_text  | `父亲。一个人出去约会...`         | Chinese transcript |
+| en_text  | `Dater. A person who...`        | English transcript |
+| category | `happy`                         | one of {angry, crying, happy, laugh, sad} |
+
+### metadata_all.tsv (overview)
+
+Same 5 columns plus one `subset` column:
+
+| column   | values | meaning |
+|----------|:------:|---------|
+| `subset` | `1hr` / `50hr` / `100hr` / `500hr` / `1000hr` | the **smallest** subset this row belongs to |
+
+Subsets are nested, so a row with `subset="1hr"` is also in 50hr, 100hr,
+500hr and 1000hr; `subset="1000hr"` means the row is only in the full
+dataset. Subset sizes (cumulative):
+
+| subset value | cumulative includes | total rows |
+|--------------|---------------------|-----------:|
+| `1hr`        | self                | 839 |
+| `50hr`       | 1hr + 50hr          | 42,467 |
+| `100hr`      | 1hr + 50hr + 100hr  | 85,004 |
+| `500hr`      | + 500hr             | 425,952 |
+| `1000hr`     | full                | 896,477 |
+
+Filter with pandas:
+
+```python
+import pandas as pd
+df = pd.read_csv("metadata_all.tsv", sep="\t")
+
+# Build any subset by including this tier and all smaller ones:
+order = ["1hr", "50hr", "100hr", "500hr", "1000hr"]
+def subset(df, target):
+    keep = order[:order.index(target) + 1]
+    return df[df["subset"].isin(keep)]
+
+df_50hr = subset(df, "50hr")   # equivalent to metadata_50hr.tsv
 ```
 
-Each pair generates two entries (EN→ZH and ZH→EN).
+## Convert to training format (s-s conversation)
+
+If your trainer expects the original `metadata_shard` format
+(`{task_type, conversation:[...]}`), use the included script:
 
 ```bash
-python make_kimi_train.py
+python scripts/tsv_to_metadata_shard.py metadata_100hr.tsv \
+  -o metadata_shard.jsonl
+# or split:
+python scripts/tsv_to_metadata_shard.py metadata_100hr.tsv \
+  --shard-size 50000 -o metadata_shard.jsonl
 ```
+
+Each TSV row produces 2 conversations (en→zh + zh→en). Use
+`--directions en2zh` / `--directions zh2en` to keep one direction.
+
+## How it was built
+
+- En/Zh paired by the `<id>_en` / `<id>_zh` suffix on the original
+  `entries_shard_*.jsonl` records (2,801 unpaired entries dropped).
+- Tier assignment per emotion: shuffle (seed=42), accumulate English
+  duration, label each pair with the smallest subset it falls into.
+- Tar shards are emotion-balanced (round-robin) within each tier and
+  capped to keep individual tars manageable (≤ ~20 GB).
+
+## Citation
+
+(TODO)
+
+## License
+
+CC BY-NC 4.0 (TODO confirm).