README.md

---
language:
- ta
task_categories:
- automatic-speech-recognition
tags:
- audio
- speech
- parquet
dataset_info:
  splits:
  - name: train
  - name: test
  features:
  - name: id
    dtype: string
  - name: audio
    dtype: audio
  - name: audio_length
    dtype: int64
  - name: caption
    dtype: string
  - name: language
    dtype: string
  - name: duration
    dtype: float64
  - name: upload_id
    dtype: string
  - name: segment_id
    dtype: string
  - name: collection
    dtype: string
  - name: label
    dtype: string
  - name: source_audio
    dtype: string
  - name: start
    dtype: float64
  - name: end
    dtype: float64
---

# Audio Clips Dataset

AudioCaps-style parquet dataset generated from `T_VOICE_OVR`.

The dataset stores clipped audio directly inside parquet shards under `data/`.
Rows are deterministically split into train/test with approximately 1% in `test`.

## Columns

- `audio`: embedded audio bytes and clip path
- `audio_length`: approximate number of samples at 16000 Hz
- `caption`: transcript text
- `language`, `duration`, `upload_id`, `segment_id`, `collection`, `label`, `source_audio`, `start`, `end`

## Usage

```python
from datasets import Audio, load_dataset

ds = load_dataset("007ask/tvoice", split="train", streaming=True)
ds = ds.cast_column("audio", Audio(sampling_rate=16000))

first = next(iter(ds))
print(first["audio"]["array"].shape, first["caption"])
```