Taiwanese Speech Utterances for Elder-Care Intent Classification

TaigiSpeech is a spoken language understanding dataset containing over 3,000 Taiwanese speech utterances from 21 speakers. Each utterance is labeled with one of 8 intent classes, designed for elder-care and smart-home voice command scenarios to support research in a low-resource language.

Use Cases

• Train an intent classification model using the 8 intent labels on over 3,000 Taiwanese speech utterances.
• Analyze speaker variation across the 21 speakers to study accent or dialect patterns in Taiwanese speech.
• Develop a speech recognition system for elder-care scenarios using the labeled voice command data.
• Build a multi-speaker model for smart-home applications using the dataset's audio and intent labels.

Strengths

• Contains over 3,000 speech utterances, providing a foundation for a low-resource language.
• Includes data from 21 different speakers, allowing for analysis of speaker variation.
• Features 8 distinct intent classes specifically designed for practical elder-care and smart-home applications.

Limitations

• The dataset size of 3,000+ utterances is relatively small for training complex speech models without augmentation.
• Limited to 8 intent classes, which may not cover the full range of commands needed for broader applications.
• Potential geographic or demographic bias is present as the data is focused on the Taiwanese language and elder-care scenarios.

Provenance

Source

TaigiSpeech organization on Hugging Face.

Collection Method

Collection of Taiwanese speech utterances from 21 speakers, labeled for intent classification.

Freshness

Last updated on 2026-03-22.

Geography

Taiwanese language focus, likely from Taiwan.