The next generation of hearables products makes new demands on embedded audio processors. They will require a multidimensional blend of digital signal processing with sophisticated neural network based algorithms delivered at ultra low energy and latency.

Our second generation GAP9 processor enables a market leading audio experience for hearable devices with features like neural network steered, ultra-low latency, active noise cancellation, neural network based noise reduction and 3D sound in hearable devices such as Truly Wireless Stereo earbuds.

The GAP9 hearable platform is exceptionally power efficient for voice and music processing. It gives a headroom in both energy and processing power that can be used to develop innovative new features with no compromise in area, cost or energy.

GAP9 also allows for multi sensor analysis (vision and sound) and is a perfect solution for battery-powered smart security systems and smart building sensors.

Our sophisticated toolset and GAP9’s inherent homogeneity and scalable performance make development significantly easier.

GAP9 product brief

GAP9 highlights

Best-in-class DSP for music and voice audio power and quality performance

State of the art programmable ultra low latency audio processing for ANC, ASRC and more

Best-in-class ultra low energy neural network performance for AI based or driven algorithms

Familiar, easy to use toolchain for embedded digital signal processing and neural network development

Key application areas

New audio experiences for hearables:

  • Active noise cancellation with neural network based acoustic scene detection
  • Noise reduction (DNN-NR)
  • Speech intelligibility
  • Voice separation
  • Immersive 3D sound
  • Spoken language understanding

Always on, battery operated inference on images, sounds and more

  • Occupancy management
  • Surveillance systems
  • Face detection / identification
  • Speaker detection / identification
  • Voice driven user interfaces

Easy development

State of the art toolchain with a wide range of supported models and frameworks

  • RISC-V C / C++ toolchain with full support for ISA extensions based on GNU toolset (GCC & GDB)
  • GAP AutoTiler code generator for explicit memory movement
  • Generator library with DSP/NN kernel support
  • GAPflow tools providing end-to-end code generation from NN frameworks such as TensorFlow and PyTorch
  • AudioTools mapping audio graphs onto hardware resources from MatLab templates
  • PULP OS, FreeRTOS™
  • Cross OS PMSIS cluster & device API
  • Debug support including on-chip debug
  • GVSOC SoC simulator and visual code Profiler