Laurent Cimon on Nostr: I just succeeded in making a start of voice commands with CMU Sphinx on my ...
I just succeeded in making a start of voice commands with CMU Sphinx on my #raspberrypi 5. I tried to adapt the en-us model but it didn’t work, so I started my own model.
It has hallucinations so I need to find a way for it not to misinterpret random talk with a voice command. I also need to find a solution to the pause command not being heard over the sound of a playing video or music.
For a single day I at least have a model that somewhat ignores misheard commands, while reliably reacting to real commands. But this is more tedious than I thought. Having training data is the hard part, and there isn’t much audio data in frenglish with Quebec’s accent.
And it’s going to get worse when I’ll work on having it recognize my girlfriend’s voice.
It has hallucinations so I need to find a way for it not to misinterpret random talk with a voice command. I also need to find a solution to the pause command not being heard over the sound of a playing video or music.
For a single day I at least have a model that somewhat ignores misheard commands, while reliably reacting to real commands. But this is more tedious than I thought. Having training data is the hard part, and there isn’t much audio data in frenglish with Quebec’s accent.
And it’s going to get worse when I’ll work on having it recognize my girlfriend’s voice.