Oronyms are words or phrases with similar pronunciations but different spellings and vastly different meanings.
Oronyms are a great basis for punning and a huge problem for speech recognition by computers, and even people (like in noisy environments).
Here’s one of the examples Linguistics Girl gives:
a nice He took a nice cold shower after his date.
an ice He took an ice cold shower after his date.
These two sentences don’t have a huge difference in meaning — both involve taking a cold shower — but this is an uncommon coincidence. More often, the pair of oronyms have very different meanings. One of the most famous examples for speech recognition is the following:
recognize speech It is very difficulty to recognize speech.
wreck a nice beach It is very difficult to wreck a nice beach.
This example demonstrates, too, how the phonology of English contributes to the oronymic qualities of the phrase. For instance, the [g] is recognize is rarely pronounced clearly in casual speech. Additionally, the [sp] sequence in speech doesn’t sound like [s]+[p], because English speakers pronounce stops like [p] without aspiration after [s] (which makes them sound like [b], the voiced and unaspirated counterpart). These two characteristics of English pronunciation lead recognize and wreck a nice to sound similar. Then, the s from speech is incorporated into nice, which leads to the remainder of the word sounding a lot like beach.
Just looking at the acoustic patterns, as computers would, makes it very difficult to tell the difference between oronyms. So how do humans do it? Well, that’s one of the interesting questions in speech perception: what tools besides the acoustic input do we use to parse the speech stream? Some of the answers linguists have found include the frequency of the words or lexical items (how often or how likely they are to occur), the part of speech that is expected next (where in the sentence you are), and the discourse environment (e.g., you don’t expect to be talking about wrecking beaches in a class on computer speech recognition).