Word/phrase-level dictionary entry recordings, likely sourced from the Ministry of Education's official Taiwanese dictionary (臺灣台語常用詞辭典) -- each entry is a single word or short phrase (not a full sentence), with Han-lo text, a Mandarin gloss, and Tai-lo romanization.

Dataset Structure

id: numeric dictionary-entry identifier (e.g. 45549), already globally unique.
audio: audio clip (16kHz, mono, 24-bit PCM WAV (majority)).
text: Taiwanese (Han-lo) dictionary entry text -- a single word or short phrase, not a full sentence (same language as the audio, per this project's field-naming convention).
mandarin: Mandarin gloss/translation of the entry.
tailo: Tai-lo (台羅) romanization of the entry.

Statistics

split	lang_name	hours	n_utts	n_chars_in_utts	secs/utt	chars/sec	n_sents	n_chars_in_sents
train	Taigi	15.8519	16859	159884	3.38	2.8	0	0
Total	-	15.8519	16859	159884	3.38	2.8	0	0

Downloads last month: 13

Use case

: ASR (Automatic Speech Recognition) / pronunciation reference

Language

: Taiwanese (Southern Min), Han-lo (漢羅) script

Source

-- neither the zip's own filename nor any object name in the bucket contains "dict"; this was only found by inspecting the zip's internal folder names directly) into HF-viewer-friendly Parquet shards.

Audio format

: 16731/16859 rows are 24-bit, 128/16859 rows are 16-bit