Skip to content

Keyword Detection (Live Audio + ML)

Real-time 'yes'/'no' keyword detection using on-board microphones and the TensorFlow Lite Micro micro_speech model. Audio is captured via I2S DMA, spectral features are extracted, and a neural network classifier runs inference every ~1 second.

Language implementations

Language Source WASM Demo
C apps/c/example_keyword_live/ Run in browser{:target="_blank"}
C++ apps/cpp/example_keyword_live/ Run in browser{:target="_blank"}
Rust apps/rust/example_keyword_live/ Not available
Zig apps/zig/example_keyword_live/ Not yet available

Key APIs demonstrated

ove_audio_graph_*, ove_audio_device_source/sink, ove_model_create, ove_model_invoke, ove_model_input/output

Required Kconfig

CONFIG_OVE_AUDIO=y, CONFIG_OVE_INFER=y

How to build

make host.posix.example_keyword_live      # C
make host.posix.example_keyword_live_cpp  # C++
make configure && make download && make && make run