Syamjith NK
syamjithnk
AI & ML interests
Arabic language technology · speech synthesis evaluation · text rendering and shaping · applied AI in film production
Recent Activity
updated a collection 3 days ago
Arabic text & speech quality new activity 3 days ago
syamjithnk/arabic-lint:Recover logical-order text with NFKC alone; real reshaper examplesOrganizations
None yet
Recover logical-order text with NFKC alone; real reshaper examples
#1 opened 3 days ago
by syamjithnk
Coverage correction: wrong-language configs, unread nested text, re-sample
#2 opened 3 days ago
by syamjithnk
Scorer: fold hamza spellings, read digit-by-digit numbers, fix لل prefix
#4 opened 3 days ago
by syamjithnk
Re-classify failure modes; the old 'ligature' verdict was a catch-all
#2 opened 3 days ago
by syamjithnk
Card: glyphs_ok caveat on 'recognisable' rows, macOS fonts, DOI label
#3 opened 3 days ago
by syamjithnk
SHA256.txt: payload only; clarify LICENSE-DATA scope
#2 opened 3 days ago
by syamjithnk
Numerals as a stress case for Arabic TTS (open data)
4
#3 opened 21 days ago
by syamjithnk
Labels are stored as presentation forms (shaping applied before saving)
1
#2 opened about 2 months ago
by syamjithnk
Arabic-Indic numerals silently break TTS: 7% vs 73% on the same sentences
2
#2 opened 2 months ago
by syamjithnk
Arabic-Indic numerals collapse to 7% while Western digits hold at 73%
2
#1 opened 2 months ago
by syamjithnk
The standard Arabic rendering recipe corrupts Arabic on any modern renderer - now filed upstream
1
#2 opened 2 months ago
by syamjithnk
The standard reshaper+bidi recipe now breaks Arabic on any HarfBuzz stack
1
#1 opened 2 months ago
by syamjithnk
Two silent Arabic rendering failures worth checking if you generate synthetic training text
#15 opened 2 months ago
by syamjithnk
The standard reshaper+bidi recipe now breaks Arabic on any HarfBuzz stack
1
#1 opened 2 months ago
by syamjithnk
Arabic-Indic numerals collapse to 7% while Western digits hold at 73%
2
#1 opened 2 months ago
by syamjithnk