Skip to Main Content
A single-chip solution for text-to-speech synthesis is presented. The chip utilizes analog nonvolatile multilevel storage (MLS) to store a corpus of natural speech elements. Synthesis is achieved by concatenating speech elements together at the word, subword, syllable, or phoneme level. An embedded micro-controller with custom peripherals performs the algorithmic tasks of the text-to-speech synthesis, namely text normalization, phonetic translation, and corpus phoneme selection. The chip is controlled via an SPI serial bus and synthesized speech is available as both analog and digital PCM audio. The chip has been fabricated in 0.5-μm flash CMOS.