|
# Dataset Description
|
|
|
|
This document provides a detailed description of the dataset's contents, structure, and the significance of each component.
|
|
|
|
## 1. Audio Recordings
|
|
The dataset includes audio recordings of participants spelling out a randomized full name, a phone number, and an address. Each participant's audio files are stored in separate folders under the `audio_data/` directory.
|
|
|
|
### Audio Files Naming Convention
|
|
- `audio_data/Names`: Audio of participants spelling out a randomized name letter by letter.
|
|
- `audio_data/Numbers`: Audio of participants reading out a randomized phone number digit by digit.
|
|
- `audio_data/Addresses`: Audio of participants stating randomized address clearly.
|
|
|
|
The folders contain raw audio files (.wav). Each participant is assigned a unique file_name, which corresponds to three specific file names in the above folders. The ground truth data, including participant names, phone numbers, and addresses, is stored in the metadata.csv file.
|
|
|
|
## 2. Metadata
|
|
The accompanying metadata file `metadata.csv` contains essential information about each participant. The columns in the metadata file include:
|
|
|
|
| Column Name | Description |
|
|
|----------------------------|-------------------------------------------------------------------------------------------------------|
|
|
| **Response_ID** | Unique identifier for each participant's response.
|
|
| **file_name** | Local file path for hugging face dataset upload compatibility. | |
|
|
| **Age** | Age of the participant in years. |
|
|
| **Gender** | Gender of the participant (e.g., Male, Female, Non-binary). |
|
|
| **Nationality** | Participant's nationality. |
|
|
| **Native Language** | The language the participant primarily speaks. |
|
|
| **Familiarity with English** | Self-reported level of familiarity with English |
|
|
| **Accent Strength (Self reported)** | Self-reported strength of the participant's accent on a scale from 0 (no noticeable accent) to 10. |
|
|
| **Difficulties** | Self-reported frequency of difficulty with automated systems |
|
|
| **Recording Machine** | Device used by the participant for recording (e.g., phone recorder, external microphone). |
|
|
| **Name** | Name recorded by the participant. |
|
|
| **Number** | Number recorded by the participant. |
|
|
| **Address** | Address recorded by the participant. |
|
|
| **Duration_secs** | Time it took to complete the survey. |
|
|
|
|
## 3. Significance of the Dataset
|
|
The dataset is crucial for:
|
|
- Reducing bias in automated speech recognition systems, particularly for non-native speakers.
|
|
- Providing researchers and developers with a resource to enhance their understanding of how different accents affect speech recognition accuracy.
|
|
- Supporting the development of more inclusive technologies.
|
|
|
|
## 4. How to Access the Data
|
|
You can access the Alphanumeric Audio Dataset in two ways:
|
|
|
|
1. Hugging Face (Recommended):
|
|
|
|
To directly load the dataset into your project using the Hugging Face datasets library, use the following Python code:
|
|
```
|
|
from datasets import load_dataset
|
|
|
|
dataset = load_dataset("sakshee05/alphanumeric-audio-dataset")
|
|
```
|
|
|
|
2. Github
|
|
Access at [alphanumeric-audio-dataset](https://github.com/Sakshee5/alphanumeric-audio-dataset)
|
|
|
