@cloudrumbles/indic-transliterate

Node.js transliteration for 21 Indian languages using AI4Bharat's IndicXlit models via ONNX Runtime.

This is an unofficial port. Models (~47MB) are bundled with the package.

Install

npm install @cloudrumbles/indic-transliterate

Usage

import { IndicTransliterator } from '@cloudrumbles/indic-transliterate'

const transliterator = new IndicTransliterator()

// Tamil
const tamil = await transliterator.transliterate('amma', 'ta')
// => ['அம்மா', 'அம்ம', 'ஆம்மா', ...]

// Hindi
const hindi = await transliterator.transliterate('namaste', 'hi')
// => ['नमस्ते', 'नमस्ती', ...]

// Get more candidates
const results = await transliterator.transliterate('chennai', 'ta', 10)

// With rescoring (improves accuracy by ~6%)
const t = new IndicTransliterator({ rescore: true })
const rescored = await t.transliterate('amma', 'ta', 4)
// => ['அம்மா', 'ஆம்மா', 'அம்ம', 'ஆமா']
// Dictionary auto-downloads on first use (~200MB per language)

API

new IndicTransliterator(options?)

Creates a new transliterator instance. Models are loaded lazily on first use.

const transliterator = new IndicTransliterator({
  beamWidth: 4,      // beam search width (default: 4)
  maxLen: 20,        // max output length (default: 20)
  rescore: true,     // use word frequency dict for better accuracy (default: false)
  rescoreAlpha: 0.9, // model vs dict weight (default: 0.9)
  modelPath: '...'   // custom model path (default: bundled)
})

transliterate(word, langCode, count?): Promise<string[]>

Transliterates a romanized word to the target script. Returns count candidates (default 5) ranked by likelihood.

getSupportedLanguages(): string[]

Returns array of supported language codes.

initialize(): Promise<void>

Pre-loads the ONNX models. Optional - models load automatically on first transliterate() call.

dispose(): Promise<void>

Releases ONNX sessions to free memory. Call when done.

isInitialized: boolean

Whether models are currently loaded.

Supported Languages

| Code | Language | Code | Language | Code | Language | |------|----------|------|----------|------|----------| | as | Assamese | hi | Hindi | ne | Nepali | | bn | Bengali | kn | Kannada | or | Odia | | brx | Bodo | ks | Kashmiri | pa | Punjabi | | gom | Konkani | mai | Maithili | sa | Sanskrit | | gu | Gujarati | ml | Malayalam | sd | Sindhi | | mr | Marathi | mni | Manipuri | si | Sinhala | | ta | Tamil | te | Telugu | ur | Urdu |

Credits

Models and vocabulary from IndicXlit by AI4Bharat (MIT License).

@article{Madhani2022AksharantarTB,
  title={Aksharantar: Towards building open transliteration tools for the next billion users},
  author={Yash Madhani and Sushane Parthan and Priyanka A. Bedekar and Ruchi Khapra and Vivek Seshadri and Anoop Kunchukuttan and Pratyush Kumar and Mitesh M. Khapra},
  journal={ArXiv},
  year={2022},
  volume={abs/2205.03018}
}

License

MIT