Siri. Alexa. Cortana. These are the names of today's most innovative digital voices used to respond to voice commands. These voices are not only processed to understand a command, but they are created through many algorithms to sound human. However, it's one thing to interpret voice signals into computerized commands, and another to create a human voice originating from a machine. Our work takes you to a set of algorithms that take a human voice and recreates the phrase in someone else’s voice. This is called voice morphing, or more commonly known as voice conversion.
The idea is simple. Person A says “What a great day!”, person B on the other hand says “I can’t wait for today to be over”. The methods and algorithms we used in this project can morph person A’s words into the mouth of person B as shown below.
By using our algorithms it’s quite possible to recreate a voice if all the voice characteristics of a person are carefully stored. This can be done by having a person read a panphonic poem like the one below.
"Please call Stella. Ask her to bring these things with her from the store: Six spoons of
fresh snow peas, five thick slabs of blue cheese, and maybe a snack for her brother
Bob. We also need a small plastic snake and a big toy frog for the kids. She can scoop
these things into three red bags, and we will go meet her Wednesday at the train station.”
- Neal Whitman
Because voice characcteristics are independent of language and are focused on the sounds, this opens up to many different types of applications such as the following.
Film dubbing: Foreign films can be spoken in the native language by converting the original voice into a foreign language using the algorithms with similar approaches and panphonic poems.
Voice Lock Security: Systems involving voice commands usesd for security purposescan be challenged by testing the authenticity of the actual person's voice opposed to an intruder.
Voice Restoration: Biomedical researchers can use algorithms similar to ours to recreate a person's voice using the methods similarly described in the film dubbing above.
Personalized Artificial Voices: Very much like Siri, we can recreate artificial voices using the approach we had for this project.
SPEECS is a student group project at the University of Michigan that aims to display small yet profound steps towards speech analysis and reconstruction. Our goal is to open possibilities for further research in areas of speech processing.
References:
http://matlabsproj.blogspot.com/2012/05/voice-conversion-in-matlab.html
http://www.fon.hum.uva.nl/praat/
http://research.cs.tamu.edu/prism/lectures/sp/l9.pdf
http://matlabsproj.blogspot.com/2012/05/voice-conversion-in-matlab.html
MIT Course ECE 345 Lecture 5
ece.ucsb.edu
Discrete Time Signal Processing Ch 13 -- Alan V Oppenheim, Ronald W Schafer
http://matlabsproj.blogspot.com/2012/05/voice-conversion-in-matlab.html
S P E E C S