Malay Speech Synthesis Dataset with 241 Hours of Audio

A collection of approximately 241 hours of high-quality Malay speech audio synthesized by the ms-MY-YasminNeural voice. The audio is split into two subsets: 99.4 hours from Malay Wikipedia and News texts, and 142 hours from Malaysian Parliament transcripts. All audio has a 24000 Hz sample rate and uses sentences between 2 and 20 words in length.

Use Cases

• Train a text-to-speech model using the 24000 Hz sample rate audio and ms-MY-YasminNeural voice characteristics.
• Fine-tune an automatic speech recognition system on the 241 hours of clean Malay audio from Wikipedia and Parliament sources.
• Analyze prosody and pronunciation patterns in Malay using sentences constrained to 2-20 words.
• Build a speech synthesis voice cloning model leveraging the super clean audio samples from the ms-MY-YasminNeural narrator.

Strengths

• Large volume of 241 total hours of synthesized speech audio.
• High-quality 24000 Hz sample rate audio described as 'super clean'.
• Text sources are clearly defined as Malay Wikipedia, News, and Malaysian Parliament transcripts.
• Sentence length is controlled, ranging from 2 to 20 words.

Limitations

• All audio is synthesized from a single neural voice (ms-MY-YasminNeural), lacking natural human speaker variation.
• The dataset consists solely of machine-generated speech, which may not capture all nuances of natural human speech.
• The temporal coverage and original recording dates of the source texts are not specified.

Provenance

Source

Texts sourced from Malay Wikipedia, News articles, and Malaysian Parliament transcripts.

Collection Method

Audio synthesized using Microsoft Azure's ms-MY-YasminNeural text-to-speech neural voice.

Freshness

Last updated on the platform in April 2022.

Geography

Malaysia, focusing on the Malay language.