Monthly Archives: April 2014

Grinding Away

There are a number of phonemes that synSinger is currently doing quite badly (D, G and CH to name a few). I’d been hoping this weekend to be able to focus on them, but the time was spent doing bug … Continue reading

Posted in Uncategorized | Leave a comment

Smaller Frames

Most Klatt-style TTS systems send 10ms-sized frames to the the synthesizer. This is feasible because speech attributes generally move fairly slowly. For synSinger, I’ve been sending frames the length of the phoneme rather than going down to a more granular … Continue reading

Posted in Uncategorized | Leave a comment

Glottal Flow Derivative

synSinger uses the Liljencrants–Fant model for generating the glottal pulse. Recently, the materials I’ve been reading have suggested that using the derivative of the glottal pulse makes for a better model. A excellent summary can be found in the Praat … Continue reading

Posted in Uncategorized | Leave a comment

Adding Stress

For a TTS (Text to Speech) application, stress indicates emphasis put on syllables, often through changes in volume and pitch. The CMU Pronouncing Dictionary conveniently tags all vowels with one of the following stress tags: 0 indicates no stress; 1 … Continue reading

Posted in Uncategorized | Leave a comment

Snap, Crackle, Pop

I’d been working on getting the “pops” out of the output. The main problem was switching the filters over from one branch and back again. For example, shifting the filters from cascade to parallel or back again would invariably cause … Continue reading

Posted in Uncategorized | Leave a comment

Klatt Resonators

I’ve changed out the filters I’ve been using for those that are specified in the Klatt synthesizer. The result is that I’m now able to put the filters in series and get usable results. For a while, I was getting … Continue reading

Posted in Uncategorized | Leave a comment

Fitting Words Into Tight Spaces

Unlike traditional TTS, a singing synthesizer needs to fit all the phonemes into a predetermined duration. For notes of short duration, this can be problematic. In the prior version of the code, the program flagged vowels (and rests) as “primary”, … Continue reading

Posted in Uncategorized | Leave a comment