I asked Argon to synthesise a voice for itself with code, no libraries, only Python.
It pulled this together, step by step:
- converted text into phonemes, stress patterns, and connected tongue-tap allophones
- built a source-filter foundation from a periodic pitch pulse and cascaded resonators
- shaped vocal fold airflow with the Liljencrants-Fant equation and open-phase breath noise
- modeled 5 vocal tract chambers, turbulent consonant hiss, and 18ms stop bursts
- connected syllables with fifth-degree minimum-jerk curves for smooth tongue and pitch motion
- added subglottal windpipe rebound, arched lung-pressure envelopes, and lip radiation physics
The video shows the evolution of the voice.
+7
