Mayan Voice: Yucatec Maya Speech Corpus for ASR Training

2,535 audio utterances in Yucatec Maya (ISO 639-3: yua) with orthographic transcriptions, compiled for training and evaluating automatic speech recognition models for low-resource indigenous languages. The corpus was created by author mau-cr as part of a thesis on MMS-based ASR for Yucatec Maya and was last updated on Hugging Face in May 2026.

Use Cases

• Training automatic speech recognition models based on the 2,535 Yucatec Maya utterances.
• Evaluating ASR model performance on a low-resource indigenous language corpus.
• Fine-tuning large multilingual speech models like MMS for Yucatec Maya.
• Studying phonetic and linguistic features of Yucatec Maya from transcribed speech data.

Strengths

• Contains 2,535 utterances, providing a foundation for model training.
• Includes audio from 32 different speakers, which likely contributes to speaker diversity.
• Audio is standardized at 16 kHz mono, which is a common format for speech processing.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset size (2,535 utterances) may be limited for training large, modern ASR models without augmentation.
• Row count and total audio duration are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Compiled for thesis work on ASR for Yucatec Maya.

Freshness

Last updated 2026-05-22 20:25:02; freshness should be verified.

Geography

Likely related to Yucatec Maya-speaking regions, but specific geographic coverage is not stated.