Rhyme search and the Proximity score

Rhyme search is the box in the toolbar above the lyrics editor. Type a word — or just the ending you want to rhyme with — and it comes back with a list of words that sound like it, the best matches first. Next to every result there is a number between 0.00 and 1.00: its Proximity.

That number is the whole point of the list. A search that only matched letters would bury the rhyme you want under a hundred words that merely look similar. Proximity is the app telling you how close each result actually sounds, so the ones worth singing sit at the top and the rest can be hidden with one drag of a slider.

Where it lives, and what comes back

The search box sits in the toolbar above the lyrics you are writing, and reads Search rhymes… until you type in it.

• The dictionary selector, immediately to its left, chooses which word lists are searched.
• ≈ turns approximate rhymes on and off, and M+ turns on multi-word rhymes — rhymes made of more than one word.
• Results open in a panel under the box, in two collapsible groups with a count on each: Unigrams for single words and Bigrams for two-word phrases. When a group holds more results than are on screen, a more link loads the next batch.
• Every row is the word on the left and its Proximity on the right. Click a row and the word is copied to your clipboard; the list stays open, so you can take several in a row.
• Across the top of the list is the Proximity row: the threshold you have set, the slider itself, and a ? button — What is proximity? — which opens this same explanation without leaving the app.

Russian, Ukrainian and English each have their own dictionaries, and the search uses the language the song is set to.

What the dictionaries do

Two different kinds of dictionary are at work here, and it pays to keep them apart.

The rhyme dictionaries are the ones behind the selector, and they decide which words can turn up at all. Russian has six — Compact, Full, Anglicisms, Surnames, Corpus Additions and Uncensored Words — plus your own User Dictionary; English and Ukrainian have one each, plus yours. The search gathers its candidates from the boxes you have ticked, and only then scores them and sorts them. Tick another box and the list gets longer. Untick them all and there is nothing to search, so nothing comes back.

The stress dictionary is a different thing entirely. The rhyme dictionaries hold spellings — where the stress falls is written in none of them. That comes from the app's own stress dictionary, plus any stresses you have set yourself, and it is what lets the app tell до́ма from дома́.

Both are needed, and this is where it is easy to guess wrong: typing a stress mark into the search box does not make the rhyme dictionaries idle. It is tempting to think it does — the mark is the precise part, so surely it takes over. It does not. The dictionaries hand over the words first; the mark then keeps the ones stressed the way yours is. It is a filter over what they offered, not a replacement for them.

The difference is not small. Searching оля́, with the mark on the last syllable, returns:

• 18 words with only Compact ticked,
• 47 words with only Full ticked,
• 510 words with all six ticked,
• nothing at all with every dictionary unticked.

Nearly 500 of those words are reachable only because more dictionaries were switched on — with the stress mark in place the whole time. So: the dictionaries set how wide your search is, the stress mark sets how precise it is. If a stressed search feels thin, the fix is usually to tick another dictionary, not to drop the mark.

What the number means

Proximity is how close a result sounds to the word you typed. 1.00 is about as close as it gets; 0.00 means the two words have no real sound in common — that one turned up because it shares some letters, not because it rhymes.

The slider sets a threshold, it doesn't search again

Every result is already in the list, best first. Dragging the slider up hides the weak matches from view; drag it back down and they all come back. No new search is ever sent.

The list shows at most a thousand rows at a time, purely so the panel stays quick, and more loads the next batch — the words past the thousandth are waiting their turn, not missing. And because the list is ordered best first, raising the slider only ever trims from the bottom: whatever is strongest is already in front of you.

The slider and the row numbers are one scale

Both run from 0.00 to 1.00, so if you spot a row at 0.70 that is about as loose as you want to go, set the slider to 0.70 and everything weaker disappears.

What makes a score high

• The stressed vowel has to match. If yours is stressed on о and the other word is stressed on у, it scores near zero however alike they look on paper — до́ма / дума is only 0.10.
• Everything after the stressed vowel has to match too. That tail is the part your ear actually hangs on.
• The ending has to be the same length. A word stressed on its last syllable will not rhyme with one stressed a syllable earlier, even when they share that vowel.
• The consonant just before the stressed vowel is a bonus. Matching it makes a Russian rhyme fuller and richer. In English it works the other way round — a match there starts to sound like the same word twice, so it counts for a little less.

What the numbers feel like

• 0.95–1.00 — perfect and rich: the whole ending lines up, and so does the consonant in front of it.
• 0.85–0.95 — a clean rhyme; the only difference is a sound you barely hear.
• 0.70–0.85 — a decent rhyme with one sound off in the ending.
• 0.40–0.70 — slant rhyme: the vowel carries it, the consonants drift apart.
• 0.10–0.40 — a faint echo, not really a rhyme.
• below 0.10 — unrelated, usually because the stressed vowel is different.

An example

Search оля. симферополя comes back at 1.00 — same stressed о, same -ля after it. мамуля comes back too, but at 0.10: it is stressed on у, so it is a near-miss the search picked up on the way. Nudge the slider to 0.20 and мамуля is gone while симферополя stays.

That is the working habit in one line: read the numbers on a few rows, decide where your own tolerance sits, and leave the slider there.

Next

A rhyme that sounds right is not always a word anyone uses. Rhyme refinement against Corpus explains the Refine button at the top of the results panel: it replaces every Proximity score with how often the word actually occurs in real texts.

Every result also arrives with an icon or two saying what kind of word it is. Part of speech — the icons on your rhymes covers all five of them, and the buttons in the Unigrams header that hide the kinds you do not need.

A rhyme that scores 1.00 still has to land on the right beat. The syllable grid is the other half of the toolkit — it shows which syllable falls on which beat, and which of them are stressed.

Or just try it: open the realyrics editor, start a song, and type an ending into the search box.