More voice, less choice: The rise of voice interfaces and the decline of open source voice
summary
Claim: by 2023 there will be about 8 billion voice assistants, but that scale comes with shrinking choice. Open projects like CMU Sphinx move commercial, Common Voice and DeepSpeech lack clear support, and over a dozen open assistants (Mycroft, Julius, Sepia, OVAL) remain fragmented. Training needs run to thousands or tens of thousands of hours of labeled audio, large compute like TPUs, and content APIs (Spotify) block access; Rwanda/Kenya models show open data can produce usable language models.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:22Kathy Reed
- 4:34Starting Points
- 6:48Claude Shannon the Father of Information Theory
- 13:03Voice Synthesis Advances
- 13:39Linear Predictive Coding
- 15:28Hidden Markov Models
- 21:54Summary
- 25:47Open Source Voice Assistants
- 34:43Conclusion
- 40:57Are There any Voice Assistants Prepared for Bilingual Users That Switch between Languages in a Single Sentence or Borrow Words from another Language Periodically
- 41:59Code Switching
- 43:33Drawback for Deep Speech
related talks
27mEnsuring Black voices matter: Why your voice assistant is racist, and what we can do about it
42mWelcome to VOICE: Opportunities and Challenges for Brands
26mEnabling the next generation in voice technology
24mOpen Source Voice AI: From ChatGPT's Infrastructure to Your App
43mAnatomy of an open source voice assistant
17mServing Voice AI at Scale
32mSpeech Technologies for a Multilingual World
40mKeynote: Open-source Multimodal AI in the Wild