DanielSWolf
commented
10 months ago Rhubarb is optimized for use in production pipelines and doesn't have any real-time support. Regarding alternatives:
- Opening the mouth based on the power of the audio signal works to a degree, but tends to look rather bad.
- Ironically, running a simple VAD to distinguish speech segments from pauses, then filling the speech segments with random mouth movements may even look better.
- Depending on your TTS system, you may be able to get precise phoneme timings without any extra work. Depending on your C++ skills, you may be able to hack Rhubarb to directly take these timings as input, skipping all the time-consuming speech recognition work.
Hi, I'm working on an app where an llm response is converted into speech using tts and the audio is played alongside an animation of a character moving its mouth. Is there a way to use your library to do this or perhaps you could point me to a better option? Another option I'm considering is to create an ascii animation and move the lips up and down based on the waveform. Do you know how I might approach this? I understand that simply looking at the crests and troughs and aligning the lips to those doesn't work. Thanks