Make Bruce Willis or Sylvester Stallone say anything
summary
An algorithm that converts a user’s voice into the voice of any chosen person using artificial neural networks is presented. The talk describes the specific neural network architecture used for voice conversion, explains why that architecture suits speech processing, touches on the underlying acoustic mechanisms, and demonstrates the system on audience-proposed sentences.
AI-written summary · about
chapters (14)
- 0:00Voice conversion intro
- 1:06Voice conversion algorithm
- 4:21Golden calf components
- 10:36Simplified voice representation
- 13:26Neural networks for conversion
- 16:59Audio sequence alignment
- 19:43Algorithm results and limits
- 22:38Potential applications
- 24:55Q&A
- 27:00Language adaptation
- 29:18Real-time conversion tools
- 31:10RNN backpropagation
- 32:34Signal simplification alternatives
- 34:18Conclusion
related talks
17mServing Voice AI at Scale
42mArtificial Intelligence That Removes Language Barriers
03mBuilding artificial people: Endless possibilities and the dark side
25mBeyond Transcription: Building Voice AI That Understands Conversations
35mTranslating Brain Electrical Signals into Actions through Models...
36mNeural Networks
21mCan AI Disrupt Speech Compression?
47mFighting AI with AI