IndicPhone

The algorithm

IndicPhone converts a word in any Indic/Brahmic script into three phonetic hashes in a single left-to-right pass. The crux of the algorithm is a universal phonetic mapping table that is shared across every script, which is used to map sounds to specific Roman characters.

1. Universal translation table

Each Brahmic script has its own distinctive 128-point Unicode block, where all blocks follow the same ISCII-derived phonetic ordering. A given offset (the position of a glyph within a block) represents the same sound in every script. For example, offset 0x15 is क in Devanagari, ক in Bengali, ಕ in Kannada, and ക in Malayalam, all the same K sound.

Blank cell = no glyph at the particular Unicode offset.

2. Vowels and modifiers

As Brahmic scripts are abugidas, a consonant carries the inherent a sound. This is assumed to be the default. Other phonetic properties are denoted by appending a specific modifier digit to the consonant's code (see §3). The broader keys are derived by dropping specific digit classes (see §3.1).

• A virama (halant) suppresses the inherent vowel and joins consonants into a conjunct. Eg: ക്ത, क्त → K0
• A visarga is dropped. Eg: കഃ, कः → K.
• A standalone vowel gets its own letter code. eg: അ, अ → A;   ഇ, इ → I.

3. Digit codes

Phonetic details beyond the mapped letter in the table is recorded by appending a modifier digit to that letter's Roman code. The finest key, key2, retains all properties; the coarser key1 and key0 are formed by stripping specific digit classes from them.

0 is not in any digit classes. 0 is not a modifier digit, but is a base sound coded after the Greek θ's "th" for the dental plosives ('dantya', eg: त थ द ध). While the anusvāra 3 is a modifier, nasalisation inherently changes a word's identity too much, so it is not discarded like other digit classes. So, both these digits are retained in every key, while every other digit class ([1], 02, 4–9) are dropped in classes to form the coarser keys.

3.1. Digit class transformation

3.1.2. Example

The algorithm parses a word character by character, left to right, in one pass. It holds at most one consonant at a time and tracks three states (idle, holding a consonant, and inside a conjunct after a virama), deciding at each character whether to emit the held consonant, merge a group, or attach a modifier. Below is an example of the state transformation with the Malayalam word നീലക്കുയിൽ (nīlakkuyil, a cuckoo).

The final key2 is then reduced to two coarser keys by stripping away classes of digits: