Compare.mp4
The above video shows Rhubarb output for a challenging synthetic voice. On the left is the result without providing any text. On the right is the Rhubarb result with source text provided.
Next Rhubarb is shown against Cherry. In this comparison Rhubarb has source text provided but Cherry just uses audio. Rhubarb struggles to get enough viseme changes to make the animation look convincing. Cherry picks more visemes but is perhaps too "chattery".
On my desktop, a HP All-in-One 24-df1xxx with an 8 core 11th Gen Intel® Core™
i5-1135G7 CPU, cheerylipsync processes 605 seconds of audio stored in .mp3
format in 13.542s (wall clock time). That is 44.7 times faster than
realtime.
rhubarb with pocketSphinx recognizer takes 9.28s (wall clock time) to
process an .ogg file of length 7.7s. That is 0.83 times faster than
realtime (i.e. slower than realtime).
rhubarb with phonetic recognizer takes 2.27s to process the same 7.7s
audio file, for a factor of 3.39 times faster than realtime. For the longer
file of 605 seconds, it takes 33.56s for a factor of 18.03 times faster
than realtime.