Datasets:
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 in multiple formats, such as .wav, .mp3, and .m4a. 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 |
|---|---|
| file_name | Unique identifier for each participant's response. |
| 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:
- 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")
- Github Access at alphanumeric-audio-dataset