Testing "Goertzel Speech"

lardom

I made a short unlisted video that shows Phi Pi's speech recognition demo. It shows my first try at making a more distinct 'template' to reduce the number of 'misses'. I tried to make my vocal inflections different.
Since the Goertzel algorithm uses trigonometry in its calculations It'll take a few days to get a good understanding of how it works. In the mean time, thanks Phil Pilgrim.

video

pedward

Does the code work on the premise of Chip's speech synthesis? You would have an FFT and look for the 4 frequency ranges. A word is matched based on a series of phonemes, and if done right you could use the same library of words for generation and detection.

Phil Pilgrim (PhiPi)

No, the Goertzel speech recognition method is not an FFT, nor does it do explicit formant extraction. It's much more ad hoc than that.

-Phil

pedward

Phil Pilgrim (PhiPi) wrote: »

No, the Goertzel speech recognition method is not an FFT, nor does it do explicit formant extraction. It's much more ad hoc than that.

-Phil

Kinda stupid to ask about FFT when it's Goertzel, but I changed what I was writing in mid thought. Though Formant extraction would be way cool.

lardom

The app was correct 100% after I changed "Left" to "Left please" and "Right" to "Turn right". It's a great piece of work and it doesn't require a specialized chip. The only miss was when I said "This is a test" which probably means a threshold has to be adjusted.
The Bresenham algorithm was easy to learn by comparison. I had no idea that processing an audio signal would involve trigonometry. I'm learning quite a bit of stuff.