Voice corpus for the Alar Kannada-English Dictionary generated using Google TTS API. Data for this corpus comes from Alar-dict/data.
The manifest.csv file serves as a structured reference to all the audio files in the dataset. It provides metadata and additional information needed for processing or analysis.
The manifest.csv file is a comma-separated values (CSV) file with the following columns:
| Column Name | Description | Example Value |
|---|---|---|
File Path |
Relative path to the audio file. | ./audio/30000-39999/35112.mp3 |
Duration |
Duration of the audio file in seconds. | 0.96 |
Word |
Text transcription or label corresponding to the audio file. | ಕನ್ನಡ |
ID |
Unique identifier for the word. | 35112 |
Head |
First character of the word. | ಕ |
There are a couple of scripts: one to generate the manifest file and the other to generate the dataset yourself.
To generate the dataset yourself, setup gcloud cli as detailed out here.
cd scripts
# Generating the manifest file
go run ./manifest
# Generate the dataset yourself
gcloud init
gcloud auth application-default login
mkdir data
curl https://raw.githubusercontent.com/alar-dict/data/refs/heads/master/alar.yml -o data/data.yml
go run ./ttsThe audio files are grouped into folders based on their ID ranges, with each folder containing files named after their unique numeric ID. For example, audio/1-9999/1234.mp3 contains the audio file for ID 1234, and 10000-19999/15000.mp3 contains the audio file for ID 15000. Each folder covers a range of 10,000 IDs.
GNU General Public License v2.0