Phonemes-to-Phrases

I used python and many of its audio modules to create an application that creates phrases from phonemes. The phrases to create were inserted into the program with an audio file. PocketSphinx then decoded the phrase to get the individual phonemes and their duration. Audio files of each individual phoneme were then downloaded from http://www.getreadingright.co.uk/phoneme-pronunciation/. Audacity was used to trim off the silence before and after the phonemes. These edited files were then inserted into python as wav file. Pyrubberband was able to set the tempo of the phoneme file so that the duration of the file would equal the duration of the phoneme in the phrase audio file. Two things were done in order to blend the phoneme audio file. Four milliseconds was then added onto the end of each phoneme. A fade in effect was also added to every phoneme along with a fade out effect for the last phoneme with pydub. The results were not perfect because the duration of the individual phoneme files was 3-4 times longer than the duration of the phoneme in the phrase file. When the tempo was changed, it skewed the voice. If a shorter audio file for the phonemes was used, the results would most likely be imporved.