scb10x/thai-dialect-isan-dataset

Dataset Card for Thai Dialect Isan Speech Corpus

Dataset Description

This dataset contains audio recordings of Isan (Northeastern Thai) speech, paired with rich transcriptions and demographic metadata. It is designed to support Automatic Speech Recognition (ASR), dialect study, and text normalization tasks for the Isan language.

The dataset features spontaneous responses to specific questions, covering two domains (General and Finance), recorded by speakers from different provinces in Northeastern Thailand.

• Language: Isan (Northeastern Thai)
• Total Examples: 10,487
• Input: Audio (WAV)
• Output: Transcriptions (Isan spelling, Thai spelling, and Raw annotated format)
• License: CC-BY-4.0

Dataset Structure

Data Splits

Split	Examples
Train	9,987
Test	500

Data Fields

Each data point contains the following fields:

• id (string): A unique identifier for the dataset entry.
• audio (audio): A dictionary containing the path to the audio file, the decoded audio array, and the sampling rate.
• raw (string): The raw transcription containing dialect-to-standard annotation tokens.
  • Format: [Isan Word]<Standard Thai Word>
  • Example: "ข้อย[กะ]<ก็>บ่ค่อยมี[แฮง]<แรง>"
• thai_spelling (string): The transcription normalized to Standard Thai spelling.
  • Example: "ข้อยก็บ่ค่อยมีแรง"
• isan_spelling (string): The transcription written in Isan spelling (phonetic to the dialect).
  • Example: "ข้อยกะบ่ค่อยมีแฮง"
• name (string): The original filename, often containing metadata codes (e.g., opentyphoon;is;x_061;gen;0049.wav).
• district (string): The district (Amphoe) where the speaker resides.
• province (string): The province (Changwat) where the speaker resides.
• age (int): The age of the speaker.
• gender (string): The gender of the speaker. Values include "m" (male), "f" (female), and "x" (not specified).
• question_id (string): The ID of the prompt question asked to the speaker.
• question (string): The text of the question asked to the speaker.
• duration (float): The duration of the audio clip in seconds.

Key Features

1. Rich Annotation Format

The raw field provides a unique mapping between Isan dialect and Standard Thai. This is valuable for dialect normalization and translation tasks.

• Format: [dialect spelling]<standard Thai spelling>
• Example: [เฮา]<เรา> indicates the speaker said "hao" (Isan) which corresponds to "rao" (Thai for "We/Us").

2. Demographic Diversity

The dataset includes speakers from multiple key provinces in the Isan region, allowing for analysis of regional accent variations. Provinces include:

• Khon Kaen
• Udon Thani
• Ubon Ratchathani
• Chaiyaphum
• Roi Et
• Maha Sarakham
• Kalasin
• Nong Bua Lam Phu
• Beung Kan

3. Prompted Speech

Recordings are responses to specific questions (found in the question field), providing context for the speech. This helps in analyzing semantic understanding and sentiment in the local dialect.

Usage

Loading the Dataset

from datasets import load_dataset
import IPython.display as ipd

# Load the dataset
dataset = load_dataset("scb10x/thai-dialect-isan-dataset")

# Select a sample
example = dataset['train'][0]

# Print transcriptions
print(f"Transcript (Isan): {example['isan_spelling']}")
print(f"Transcript (Thai): {example['thai_spelling']}")

# Listen to audio
ipd.Audio(example['audio']['array'], rate=example['audio']['sampling_rate'])