---
dataset_info:
- config_name: audio
  features:
  - name: audio
    dtype: audio
  - name: file_name
    dtype: string
  - name: transcript
    dtype: string
  - name: speakerID
    dtype: string
  - name: language
    dtype: string
  - name: gender
    dtype: string
  - name: state
    dtype: string
  - name: district
    dtype: string
  - name: pincode
    dtype: string
  - name: duration
    dtype: float32
  - name: languagesKnown
    dtype: string
  - name: stay_years
    dtype: string
  - name: isTranscriptionAvailable
    dtype: string
  - name: referenceImage
    dtype: string
  - name: speakerImageHash
    dtype: string
  - name: UtteranceSequenceID
    dtype: int32
  splits:
  - name: train
    num_bytes: 218646505
    num_examples: 1302
  download_size: 214669976
  dataset_size: 218646505
- config_name: images
  features:
  - name: image
    dtype: image
  - name: file_name
    dtype: string
  splits:
  - name: train
    num_bytes: 160409135
    num_examples: 973
  download_size: 160426214
  dataset_size: 160409135
configs:
- config_name: audio
  data_files:
  - split: train
    path: audio/train-*
- config_name: images
  data_files:
  - split: train
    path: images/train-*
---
# Vaani Sample Data — Andhra Pradesh / Annamaya

A small sample slice of the [ARTPARK-IISc/VAANI](https://huggingface.co/datasets/ARTPARK-IISc/VAANI) dataset, covering the `AndhraPradesh_Annamaya` subset. 
## Configs

| Config   | Split | Rows | Description |
|----------|-------|-----:|-------------|
| `audio`  | train | 1,302 | Telugu speech utterances with transcripts and speaker metadata. |
| `images` | train |   973 | Reference images cited via `referenceImage` in the audio config (973 unique images across the 1,302 audio rows). |

## Loading

```python
from datasets import load_dataset

audio = load_dataset("SujithPulikodan/Vaani-sample-data", "audio", split="train")
images = load_dataset("SujithPulikodan/Vaani-sample-data", "images", split="train")
```

## Schema

### `audio`

| Column | Type | Notes |
|---|---|---|
| `audio` | `Audio` | Embedded audio (WAV bytes + path). |
| `file_name` | `string` | Audio basename. |
| `transcript` | `string` | Telugu transcription. May contain `<noise>` markers. |
| `speakerID` | `string` | Stable speaker identifier. |
| `language` | `string` | Spoken language (Telugu). |
| `gender` | `string` | Speaker gender. |
| `state`, `district`, `pincode` | `string` | Recording location. |
| `duration` | `float32` | Seconds. |
| `languagesKnown` | `string` | Stringified list, e.g. `['Telugu']`. |
| `stay_years` | `string` | How long the speaker has lived in the district. |
| `isTranscriptionAvailable` | `string` | `Yes`/`No` (always `Yes` after filtering). |
| `referenceImage` | `string` | Path of the prompt image shown to the speaker (matches a row in the `images` config). |
| `speakerImageHash` | `string` | Hash of the speaker's image. |
| `UtteranceSequenceID` | `int32` | Sequence index within a session. |

### `images`

| Column | Type | Notes |
|---|---|---|
| `image` | `Image` | JPEG bytes. |
| `file_name` | `string` | Image basename, matches the basename in `audio.referenceImage`. |

To join an audio row with its image:

```python
import os
imgs_by_name = {row["file_name"]: row["image"] for row in images}
def image_for(audio_row):
    return imgs_by_name[os.path.basename(audio_row["referenceImage"])]
```

## Source & License

Derived from **ARTPARK-IISc/VAANI** by IISc / ARTPARK. Refer to the upstream dataset card for the original license and terms of use; this redistribution inherits them. Please cite the upstream project when using this data.