aeneas

Why we included this project

aeneas takes an audio recording of someone reading a text and produces a timestamp map that links each line or phrase to the exact moment it is spoken, a task known as forced alignment. It matches speech recognition features (MFCC) against the text with dynamic time warping, and when no transcript alignment exists, it synthesizes the reference audio using a TTS engine. It can export SRT and VTT captions, SMIL for EPUB 3 read-along ebooks, and TextGrid and other formats used in linguistic research. Anyone building audiobook apps, karaoke-style readers, subtitle generators, or spoken-language datasets can wire the CLI or Python API into their pipeline. The tool handles 38 languages and batch jobs, and the docs walk through everything from a single file up to multi-level word and sentence alignment.