Update Gametime (basic) Parquet data

README.md

@@ -0,0 +1,122 @@
+---
+# Data-files config (no script, Parquet-only)
+configs:
+- config_name: basic
+  data_files:
+  - split: train
+    path:
+      - "basic/train-*.parquet"
+      - "basic/train.parquet"
+  - split: test
+    path:
+      - "basic/test-*.parquet"
+      - "basic/test.parquet"
+  default: true
+
+pretty_name: "Gametime (Basic)"
+tags:
+  - audio
+  - speech
+  - tts
+  - asr
+  - benchmark
+task_categories:
+  - automatic-speech-recognition
+  - text-to-speech
+language:
+  - en
+license: cc-by-4.0
+size_categories:
+  - n<100K
+---
+
+# Gametime — Basic (Parquet)
+
+**Gametime** is a lightweight speech benchmark for quick TTS ↔︎ ASR sanity checks on short, instruction-like utterances.
+
+This repo uses **Parquet-only** data files (no `Audio` feature, no WAVs checked in). Each example contains:
+- `id, category, dataset, split, template_idx, item_idx, text`
+- `audio_bytes` (raw file bytes), `audio_format` (e.g., `"wav"`), `sampling_rate` (e.g., `16000`)
+
+Why? This keeps streaming reliable and dependency-light.
+
+---
+
+## TL;DR — Stream and decode
+
+```python
+from datasets import load_dataset
+import io, soundfile as sf
+
+ds = load_dataset("gametime-benchmark/gametime", "basic", split="train", streaming=True)
+
+ex = next(iter(ds))
+buf = io.BytesIO(ex["audio_bytes"])          # raw bytes from Parquet
+wav, sr = sf.read(buf, dtype="float32")      # decodes in-memory
+print(ex["id"], sr, len(wav), ex["text"])
+````
+
+* Works with **streaming=True** (no full download).
+* No `torchcodec` required. You only need `soundfile` (libsndfile).
+
+---
+
+## Schema
+
+Columns in each Parquet row:
+
+* `id`: unique identifier (e.g., `1-a-Sequence-Number/train/1-a-Sequence-Number-01-01.wav`)
+* `category`: `"basic"`
+* `dataset`: group name (e.g., `1-a-Sequence-Number`)
+* `split`: `train` or `test`
+* `template_idx`, `item_idx`: indices if applicable (empty string otherwise)
+* `text`: reference text to speak/recognize
+* `audio_bytes`: raw WAV bytes
+* `audio_format`: `"wav"`
+* `sampling_rate`: e.g., `16000`
+
+---
+
+## Example: quick ASR sanity check (Whisper)
+
+```python
+# pip install transformers datasets soundfile evaluate
+import io, soundfile as sf, evaluate
+from datasets import load_dataset
+from transformers import pipeline
+
+asr = pipeline("automatic-speech-recognition", model="openai/whisper-small")
+wer = evaluate.load("wer")
+
+ds = load_dataset("gametime-benchmark/gametime", "basic", split="test", streaming=True)
+
+refs, hyps = [], []
+for i, ex in enumerate(ds):
+    wav, sr = sf.read(io.BytesIO(ex["audio_bytes"]), dtype="float32")
+    hyp = asr({"array": wav, "sampling_rate": sr})["text"]
+    refs.append(ex["text"]); hyps.append(hyp)
+    if i >= 199:  # quick 200-sample check
+        break
+
+print("WER:", wer.compute(references=refs, predictions=hyps))
+```
+
+---
+
+## Notes
+
+* **Advanced** category (with `<sil="Xms"/>` timing tokens) is planned separately.
+* License: **CC BY 4.0** for text, metadata, and audio.
+
+---
+
+## Citation
+
+```
+@dataset{gametime_basic_2025,
+  title   = {Gametime: Basic Speech Benchmark},
+  year    = {2025},
+  url     = {https://huggingface.co/datasets/gametime-benchmark/gametime},
+  note    = {Parquet-only split "basic" for TTS/ASR sanity checks}
+}
+```

basic/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e255c2fd8b2b33be1714b888849a985f75bd8e195a4efd8bb645d68bebdffe1
+size 240085790

basic/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:283c7ce2dc1833ecb0b82fe2c45f943ab9727a27ee23a325ebc52f80d7acb247
+size 473425907