Skip to content

Cynthia

Cynthia is the speech kernel of Assistivox: everything Assistivox says goes through it. Bill Hollingsworth wrote the Cynthia speech engine in graduate school, from 2003 to 2005, with its own pronouncing alphabet, syllable parser, phonological rules, and models of timing and intonation. Assistivox continues it.

  1. Names keys and characters. Key echo, menus and spelling get each key’s name from Cynthia: “ay” for a, “space bar”, “period”.
  2. Pronounces words. Cynthia looks each word up in its lexicon, written in cpbet (Writing Pronunciations), and then in its dictionaries.
  3. Works in its own phones. Every pronunciation becomes Cynthia’s phones, which carry the features of Cynthia’s alphabet (voicing, place and manner, height, length). Cynthia’s syllable parser divides them into onset, nucleus and coda.
  4. Gives each voice what it takes. Kokoro and Piper learned different transcription systems, and Piper voices differ from each other. Cynthia turns its phones into the symbols each voice was trained on.
  5. Changes the speed. Each voice speaks at its own best speed; Cynthia then makes the audio faster or slower with the pitch kept (WSOLA, waveform-similarity overlap-add).