I am trying to create a lightweight speech synthesizer for diphones. Everything seems pretty simple, because my native language has pretty simple rules for writing and processing text. The only problem I encountered is the step control.
As I understand it, most voice synthesizers use LPC (linear predictive coding) to control the pitch, which essentially separates the pitch information from the recorded voice samples, and then during the synthesis I can provide my own pitch if necessary.
The problem is that I'm not a DSP specialist. I used the Ooura FFT library to extract AFR information, I know a little about the use of Hann and Hamming windows (implemented the C ++ code itself), but mostly I consider DSP algorithms as black boxes.
I was hoping to find some kind of open source library which is simple LPC code with usage examples, but I could not find anything. Most of the available code (for example, the festival engine) is closely integrated into the synthesizer, and it would be rather difficult to separate it and find out how to use it.
Is there an open source DSP library C / C ++ / C # / Java with a black box LPC algorithm and usage examples, so I can just throw PCM data samples on it and get the LPC encoded output, and then drop the encoded data and synthesize decoded speech data?
source
share