MinSpeech: Cleaned Multi-dialect Min-nan Speech Corpus for ASR

MinSpeech is a cleaned, multi-dialect Min-nan (Southern Min) speech dataset maintained as a private research fork by user 'scbz'. The dataset is intended for fine-tuning Automatic Speech Recognition and Speech-to-Text Translation models. It was last updated on the Hugging Face platform on April 20, 2026.

Use Cases

• Fine-tuning Automatic Speech Recognition (ASR) models based on the described multi-dialect Min-nan audio.
• Training Speech-to-Text Translation (S2TT) systems based on the cleaned speech corpus.
• Researching dialectal variation and robustness in speech AI models based on the multi-dialect nature of the data.

Strengths

• Data has been cleaned, which likely improves signal-to-noise ratio and transcription accuracy.
• Corpus covers multiple dialects of Min-nan, potentially increasing model robustness.
• Dataset is actively maintained, with a last update recorded on 2026-04-20.

Limitations

• Dataset is listed as private, which restricts access and reproducibility.
• Column-level documentation, sample data, and dataset size are unknown, limiting suitability assessment.
• License information is unknown, creating legal uncertainty for use and redistribution.

Provenance

Source

Derived from the MinSpeech corpus; maintained by user 'scbz' on Hugging Face.

Collection Method

Likely involves cleaning and processing of an existing speech corpus.

Time Range

null

Freshness

Last updated 2026-04-20 04:16:28.

Geography

null