sopheakchan/khmer_asr_dataset

Overview

This dataset contains 19,907 Khmer speech samples prepared for Automatic Speech Recognition (ASR). The dataset is a merged collection from three publicly available sources, with additional preprocessing and manual transcript adjustments.

Disclaimer: I do not own the original audio data included in this dataset. All rights and credits belong to the respective original authors and data collectors listed below

• Seanghay

  Khmer MPWT Speech Source: https://huggingface.co/datasets/seanghay/khmer_mpwt_speech

  Khmer Speech Corpus Source: https://huggingface.co/datasets/seanghay/km-speech-corpus

• Google LLC OpenSLR SLR42

  Khmer Male Speech Dataset Source: https://www.openslr.org/42/

Column	Description
audio	16kHz speech audio
transcription	Original transcript from source
transcript_adjusted	Cleaned and manually reviewed transcript
id	Unique identifier

Special character removal ! " # $ % ' + , - . / : ? \ …

Invisible characters removed Zero-width space (\u200b)

Number normalization Latin digits were converted to Khmer digits. 123 → ១២៣

Symbol replacements (selected cases) Some contextual replacements were applied when appropriate, for example: $ → ដុល្លារ

Note: Not all symbols were replaced contextually. Most were removed during cleaning and even after manual checking, some transcription errors may still exist. Further verification is recommended