Phonemes: the building blocks of spoken language

Updated October 31, 2026

Every spoken language is built from a small set of sounds called phonemes. Understanding them explains why accents exist, why spelling is often unreliable, and how a computer knows how to pronounce what you type.

What is a phoneme?

A phoneme is the smallest unit of sound that can change the meaning of a word. In English, "pat" and "bat" differ by one phoneme, /p/ versus /b/. So do "ship" and "sheep", which differ only in the vowel.

Phonemes are about sound, not letters. "Cat" and "kite" start with the same phoneme, /k/, written with different letters. The letters "th" in "thin" and "this" are the same, but they stand for two different phonemes.

Languages use different sets of sounds

Each language picks its own set of phonemes from the many sounds the human voice can make. The numbers vary a lot. Most varieties of English have roughly 44 phonemes, depending on the accent. Hawaiian has one of the smallest inventories, with around 13. Some languages of southern Africa have well over 100, including click consonants.

This is why learners find certain sounds hard. Japanese does not distinguish /r/ and /l/, so English speakers' "rice" and "lice" can sound the same to a Japanese learner. Many English speakers struggle with the French "u" in "tu", or the rolled "r" in Spanish, because English doesn't use those sounds to tell words apart.

Spelling and sound don't always match

Some languages have very regular spelling: in Finnish, Spanish or Turkish, you can usually pronounce a word correctly just by reading it. Others, like English and French, have spelling that reflects history more than current pronunciation. Think of "though", "through", "tough" and "cough".

For text to speech, regular spelling makes life easy. Irregular spelling means the system needs a large pronunciation dictionary plus clever guessing for words that aren't in it.

The International Phonetic Alphabet

Because spelling is unreliable, linguists use the International Phonetic Alphabet (IPA), which has one symbol for each sound, whatever the language. For example:

• "ship" is /ʃɪp/ and "sheep" is /ʃiːp/
• "thin" is /θɪn/ and "this" is /ðɪs/
• French "tu" is /ty/

Dictionaries use IPA to show pronunciation. Speech engines use IPA or similar phoneme codes internally.

How text to speech uses phonemes

When you type a sentence, a speech engine first converts the words into phonemes, a step called grapheme-to-phoneme conversion. For example, eSpeak NG has hand-written pronunciation rules and dictionaries for each language. Piper uses eSpeak NG's phoneme conversion and then a neural network to turn those phonemes into a natural voice. The neural model then works out how each sound blends into the next, since real speech is continuous, not a string of separate sounds.

This also explains why choosing the right voice for your language matters. A voice trained on English only knows English phonemes and English rhythm. Give it German text and it will try to read it as English.

Stress, tone and length

Phonemes aren't the whole story. In many languages, stress changes meaning: "REcord" (a noun) and "reCORD" (a verb). In tonal languages such as Mandarin, Vietnamese, Thai and Yoruba, the pitch of a syllable changes the word entirely. Some languages, like Finnish and Japanese, distinguish short and long sounds. A good speech engine has to get all of these right. Read more about rhythm and melody in speed, pitch and rhythm.