Khmer Speech Corpus With 10.4 Hours of Audio

10.4 hours of Khmer speech audio with a mean duration of 2.5 seconds per sample, compiled by author seanghay and last updated in May 2023. It contains audio clips ranging from 0.45 to 19.39 seconds, sampled at 16 kHz. The dataset is hosted on Hugging Face and is associated with text-to-speech and automatic speech recognition tasks.

Use Cases

• Train text-to-speech models using paired audio clips and corresponding Khmer text transcripts.
• Develop automatic speech recognition systems for Khmer by learning from audio features and phonetic transcriptions.
• Analyze speech characteristics like duration and amplitude across 10.4 hours of Khmer audio data.
• Benchmark model performance on Khmer speech tasks using the provided sampling rate of 16000 Hz.
• Study phonetic patterns in Khmer language using the distribution of audio clip lengths from 0.45 to 19.39 seconds.

Strengths

• Total duration of 10.4 hours provides a substantial audio corpus for model training.
• Clear audio specifications with a consistent sampling rate of 16000 Hz.
• Detailed metadata includes min, max, and mean audio durations (0.45s, 19.39s, 2.51s).

Limitations

• Unknown total number of individual audio samples or rows.
• Potential bias from unknown speaker demographics and recording conditions.
• Lack of column details prevents assessment of data structure and available features.

Provenance

Source

Hugging Face dataset repository by author seanghay.

Freshness

Last updated in May 2023.

Geography

Likely focused on Khmer-speaking regions, but specific coverage is unknown.