Project Voice - Building Voices for All
summary
An open dataset and models let communities record and validate speech to train open source speech recognition using a DeepSpeech-derived stack and crowdsourced audio. Project Common Voice collects 5,000+ sentence sets, targets 10,000 hours of English training data, stores ~25 GB per 1,000 hours, and already mobilizes 11,000 contributors across 16–17 languages. Recordings get multiple validations, feed the neural net, and produce offline-capable models for phones and apps while campaigns continue to reach scale.
AI-written summary · about
chapters (4)
related talks
44mProject CapisciAMe
02h 04mMainstage Opening Session
30mBuilding a Sign-to-Speech Prototype with TensorFlow, Pytorch, and DeepStack
12mKeynote
23mDesigning for Accessibility
18mFacilitating Development of Pragmatic Competence through a Voice-driven Video Learning Interface
14mAI for a Billion Voices: Taking AI where there is Maximum Impact
46mArtificial intelligence to the rescue of accessibility