710765AJA.LA STUDIOS: A Novel System for Recognizing and Synthesizing Speech for African LanguagesClosedGRD Proof of ConceptInnovate UKhis project will demonstrate commercial viability of automated voice-recognition (AVR) and text-to-speech (TTS) for Sub-Saharan African languages. AVR receives speech and applies a mathematical model that converts the speech into information a computer can understand and manipulate. TTS decomposes text into a symbolic representation a computer can understand and applies a statistical model that manipulates the symbols to synthesize speech humans can understand. Commercial AVR and TTS systems are available for languages of Europe, the Americas, developed Asia, and China. Globally, speech products represent a €20b industry and have applications in industries including financial services, health care and telecoms. AVR and TTS vendors have mostly limited commercial coverage of Sub-Saharan Africa to South Africa. Growing regional economies and an expanding consumer class suggest a significant commercial opportunity for speech products across Sub-Saharan Africa. Key challenges associated with AVR and TTS for African languages include language tonality, ligatures in colloquial speech, and unique phonemic combinations. Many African languages have limited documentation and offer many dialects. AVR and TTS systems should also be accessible to the widest audience possible, with consideration for regional connectivity & device demographics. This project addresses three key innovations. Firstly, design of a mathematical model to recognize speech at commercially viable qualitative & quantitative error rates. Secondly, design of a statistical model to synthesize speech from text. Both models will address the unique challenges described earlier. Thirdly, design a system for delivering AVR and TTS in low connectivity areas and on older mobile device models to capture the widest possible audience. For tractability and to address two key markets of economic and demographic significance in Sub-Saharan Africa, the project will target a small vocabulary of Yoruba and Swahili words.