audio-noise-suppression

[UNTESTED] — v0.5 paper-correct

Builds clean on native_sim/native/64. Real-silicon dispatch + the spectral chain on V2N M33 land in v0.6 HiL.

Real-time noise suppression demo on the V2N (and V2H by SKU flip). Target market: USB-C headsets, true-wireless hearables, desktop conferencing rigs — anywhere you want cloud-grade denoising at sub-perceptual latency without shipping audio off-device.

Source: examples/audio-noise-suppression/.

Pipeline

TAS2563 mic ──▶ <alp/audio.h> ──▶ <alp/dsp.h> ──▶ <alp/inference.h> ──▶ <alp/audio.h> ──▶ TAS2563 spk
(I2S0 RX,         240 fr / 10 ms     Hann + 256-pt    RNNoise-style int8    I2S0 TX
 24 kHz S16)      blocks             FFT → mag bins   CNN → gain mask

The spectral chain routes through whichever backend the SoM has — pure C on the V2N M33, CMSIS-DSP on AEN.

board.yaml

schema_version: 2

som:
  sku: E1M-V2N101

carrier:
  name: E1M-X-EVK

cores:
  m33_sm:
    os: zephyr
    app: ./src
    peripherals: [i2s, i2c]
    libraries: [cmsis-dsp]
    inference: { backend: cpu, tensor_arena_kb: 96 }

chips:
  - tas2563

diagnostics:
  log_level: info

See also

• <alp/audio.h> · <alp/dsp.h> · <alp/inference.h>
• Examples overview