Decoder. plain-English AI glossary

Phoneme

● Core

The smallest unit of sound in language that distinguishes meaning - like /p/ and /b/ in "pat" vs. "bat."

Think of it like

In English, letters; in speech, phonemes. Phones (sounds) are instances; phonemes are categories.

Example

English: "cat" = /k/ /a/ /t/ (3 phonemes). "ship" = /sh/ /i/ /p/ (3 phonemes). "thought" = /th/ /o/ /t/ (3 phonemes).

How it actually works

Language-specific inventory. English ~= 44 phonemes; Spanish ~= 24. Recognition: acoustic models predict phonemes from spectrograms (Gaussian mixtures, DNNs, HMMs). Text-to-speech needs phoneme-level alignment. CTC loss (Connectionist Temporal Classification) allows direct acoustic-to-phoneme without explicit alignment.

For product teams

Core for speech recognition, TTS, and language learning apps.

For engineers

Datasets: TIMIT (phoneme labels). Metrics: phoneme error rate (PER). Models: acoustic models trained with HMMs or CTC.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome