Data sources

Where the readings come from, and under what terms.

Wiktionary

Headwords and alternative readings come from the Thai G2P Wiktionary corpus published by PyThaiNLP, scraped from Wiktionary in July 2022. Licensed under CC BY-SA 3.0. The dictionary file this app ships is a derived work and is offered under CC BY-SA 4.0, a later version that CC BY-SA 3.0 permits.

github.com/PyThaiNLP/thai-g2p-wiktionary-corpus

Thai Language Toolkit (tltk)

The primary reading of each headword was generated with tltk, the Thai Language Toolkit by Wirote Aroonmanakun (Chulalongkorn University), released under the BSD 3-Clause License. tltk runs only when the dictionary is built; none of its own data is included in this app.

pypi.org/project/tltk/

Reference word list

Accuracy is measured against 2,641 words with RTGS readings from our sister project thai-typing. Known errors in that list are corrected with a note on why.

UD_Thai-PUD (Universal Dependencies) — CC BY-SA 3.0
github.com/UniversalDependencies/UD_Thai-PUD

Volubilis 23.1 (Francis Bastien) — CC BY-SA 4.0
belisan-volubilis.blogspot.com/

Handwriting fonts

The handwriting recognizer was trained on synthetic samples rendered with 63 Thai fonts: 33 from Google Fonts (SIL Open Font License 1.1), 9 from Fonts-TLWG (GNU GPL 2.0 or later with the font-embedding exception; Waree under the Bitstream Vera licence), BoonJot (SIL OFL 1.1), 16 iannnnn faces from f0nt.com (f0nt.com licence: free for personal and commercial use, no redistribution of the font files) and 4 original SIPA national fonts (2007 licence agreement). The fonts were used only to render training images; no font file is included in this app.

Charm, Itim, Kodchasan, Mali, Srisakdi, Bai Jamjuree, Chakra Petch, Charmonman, Fahkwang, IBM Plex Sans Thai Looped, K2D, KoHo, Maitree, Niramit, Noto Sans Thai Looped, Noto Serif Thai, Pridi, Sarabun, Taviraj, Thasadith, Trirong, Krub, Playpen Sans Thai, Chonburi, Sriracha, Anuphan, Athiti, IBM Plex Sans Thai, Kanit, Mitr, Noto Sans Thai, Pattaya, Prompt — SIL OFL 1.1
fonts.google.com/?subset=thai

Fonts-TLWG 0.7.4: Purisa, Garuda, Kinnari, Loma, Norasi, Sawasdee, Umpush, Laksaman — GNU GPL 2.0+ with font exception; Waree — Bitstream Vera
github.com/tlwg/fonts-tlwg

BoonJot 2.0 — SIL OFL 1.1
github.com/fontuni/boonjot

iannnnn COW, DOG, DUCK, GOOSE, HEN, MONKEY, OWL, RABBIT, TIGER, PDF 2008, DVD, HBO, JPG, MTV, SOS, UBC (Prachya Singhto) — f0nt.com licence
www.f0nt.com/about/license/

TH Charmonman, TH Srisakdi, TH Charm of AU, TH Mali Grade 6 (DIP / SIPA, 2007) — SIPA font licence agreement
www.f0nt.com/release/13-free-fonts-from-sipa/

SIL Open Font License 1.1
openfontlicense.org/

Handwriting data

The recognizer is also trained on the character set of Burapha-TH (Onuean et al., Applied Sciences 2022): scanned handwriting of 68 Thai characters on 1,153 forms (1,072 writers according to the paper), published by Burapha University without an explicit licence and described by its authors as free for other use. Accuracy is measured against the KVIS Thai OCR Dataset (Joseph & Anantaprayoon, 2018): the 44 Thai consonants by 27 writers, CC BY 4.0, used only for evaluation. Neither dataset is included in this app; only the trained weights are.

Burapha-TH — Onuean, A. et al., Applied Sciences 2022, 12(8):4083, doi:10.3390/app12084083 (Burapha University / KISTI)
services.informatics.buu.ac.th/datasets/Burapha-TH/

KVIS Thai OCR Dataset — Joseph, R. & Anantaprayoon, P., Mendeley Data V1 (2018), doi:10.17632/8nr3pbdk5c.1 — CC BY 4.0
data.mendeley.com/datasets/8nr3pbdk5c/1

KVIS Thai OCR Dataset — CC BY 4.0
creativecommons.org/licenses/by/4.0/

Licenses

Dictionary data: CC BY-SA 4.0. tltk: BSD 3-Clause. Handwriting fonts: SIL Open Font License 1.1, GNU GPL 2.0+ with font exception, Bitstream Vera, and the f0nt.com and SIPA licences (used only to render training images). Burapha-TH training data: no explicit licence, cited as its authors ask. KVIS evaluation data: CC BY 4.0. The app itself is separate from the data and is not covered by these licenses.

辞書データ — CC BY-SA 4.0
creativecommons.org/licenses/by-sa/4.0/

Thai-PUD, Wiktionary 2022 corpus — CC BY-SA 3.0
creativecommons.org/licenses/by-sa/3.0/

tltk — BSD 3-Clause
opensource.org/license/bsd-3-clause

TNC 頻度 — CC0 1.0
creativecommons.org/publicdomain/zero/1.0/