Varna

Varna (Sanskrit: वर्ण — letter, character, sound) — multilingual language engine for AGNOS

Structured, queryable corpus of human language data. Phoneme inventories, writing system metadata, grammar profiles, and lexicon access for 50+ languages.

Used by shabda (G2P conversion), shabdakosh (pronunciation dictionary), svara (vocal synthesis), jnana (knowledge system), and vidya (programming reference).

Modules

Module	Description
`phoneme`	IPA phoneme inventories per language, articulatory features (manner, place, voicing), stress/tone patterns
`script`	Writing system metadata: alphabet, syllabary, logographic, abjad, abugida. Unicode ranges, directionality
`grammar`	Morphological typology (isolating, agglutinative, fusional, polysynthetic), word order, case systems
`lexicon`	Core vocabulary per language (Swadesh lists, frequency-ranked word lists), cognate detection

Features

Feature	Default	Description
`std`	yes	Standard library support
`logging`	no	Structured logging via `LIPI_LOG` env var
`full`	--	Enables all features

Quick Start

[dependencies]
varna = "0.1"

use varna::phoneme::{self, StressPattern};

// Get the English phoneme inventory
let en = phoneme::english();
assert!(en.has("θ"));           // voiceless dental fricative (think)
assert!(en.has("ð"));           // voiced dental fricative (this)
assert!(!en.has("ʀ"));          // no uvular trill in English
assert_eq!(en.stress, StressPattern::Free);
println!("{} consonants, {} vowels", en.consonant_count(), en.vowel_count());

// Look up a specific phoneme
let sh = en.find("ʃ").unwrap(); // postalveolar fricative (ship)

Architecture

varna (this) — language structure & phoneme inventories
  | provides phoneme sets per language
shabda — G2P conversion (currently English-only, varna makes it multilingual)
  | produces phoneme sequences
shabdakosh — pronunciation dictionary (currently CMUdict, varna adds IPA dicts)
  | lookup fallback
svara — vocal synthesis (consumes phonemes, produces audio)
  | voice output
dhvani — audio engine (mixing, DSP, output)

Also feeds:

jnana — multilingual knowledge access
vidya — programming concepts explained in native languages
vansh (planned) — voice assistant with multilingual TTS/STT
sahifa (planned) — OCR language detection, multilingual document processing

Development

make check     # fmt + clippy + test + audit
make bench     # Run benchmarks with history tracking
make coverage  # Generate coverage report
make doc       # Build documentation

License

GPL-3.0-only. See LICENSE.