Act on this, mostly facts

Alkove: a 7.5 MB on-device voice front-end with typed wake words

What to do

Try the Apache-2.0 early preview if your local assistant needs speaker-only audio before STT.

Why

It has npm, pip, Android and LiveKit or Pipecat packages, and keeps only the enrolled voice.

What the research found

Alkove is an early-preview Apache-2.0 on-device voice front-end of five small models (about 7.5 MB total) that turns a typed wake word into pronunciations and keeps only the enrolled speaker's voice before STT, with npm, pip, Android and LiveKit/Pipecat packages.

Claims and how well the evidence supports them

Alkove consists of five on-device models totaling about 7.5 MB 97%
Code and weights are released under Apache-2.0 99%
Focus runs at 38.6 ms per second of audio on one phone thread natively 99%
Packages are available on npm, pip, Android Kotlin, and LiveKit and Pipecat plugins 98%

Evidence

reddit.com https://reddit.com/r/LocalLLaMA/comments/1xbmhct/i_built_a_small_ondevice_voice_frontend_for_local
github.com https://github.com/alkove-ai/alkove

Actionability rated medium.