Speech, and the tools around it.

Lately I work on three things:

• Keyword spotting: detecting a wake word while keeping false wake-ups low.
• Speech enhancement: recovering clean speech from noise and reverberation.
• Speech coding: neural codecs that carry speech at low bitrates.

The projects are tools I wrote for my own work and open-sourced. tea keeps a Mac running as a server with the lid closed.

Teams can subscribe to mainstream models and a speech API; see for teams.

Write to hello@soundadam.com, or read the code on GitHub.