Skip to main content
The audio subsystem powered by low-power AI (LPAI) delivers voice user interface and audio experiences. It uses a dedicated hardware-based AI accelerator for machine learning-based work.
Audio component overview

Audio component overview

The audio system includes:
  • Application processor – CPU that handles audio processing tasks. Tasks include:
    • Managing audio record and playback
    • Decoding audio formats
    • Using LPAI for postprocessing tasks
  • Low-Power AI (LPAI) – Subsystem that runs audio playback/record and voice-activation (VA) algorithms. It integrates with a dedicated Qualcomm® Hexagon™ Processor (QDSP6) and a low-power island (LPI).
  • Audio codec – Hardware that includes:
    • Analog-to-Digital Converter (ADC)
    • Digital-to-Analog Converter (DAC) These convert analog audio to digital, and vice versa.
  • Speaker AMP and microphone – Devices that connect over I2S/TDM/SoundWire.

​
Architecture

The following figure shows the high-level audio software architecture.
High-level audio software architecture

High-level audio software architecture

LegendApplication layerMiddleware layerHardware abstraction layerKernel spaceLPAIPCPipeWire PulsePALARGSAGMTinyALSAApplicationACDBQACTASOCSPFCodecUSBQualcommOpen sourceHardwareData flowData + controlControlPipeWire PAL pluginGStreamerPipeWire High-level audio software architecture The following are the major audio software architecture components: PipeWire PipeWire is a multimedia server for Linux that routes audio between Apps and hardware. It replaces PulseAudio, offering low-latency, secure, and flexible media handling for applications. Platform Abstraction Layer (PAL) Provides higher-level audio-specific APIs to access the audio hardware and drivers to enable audio features. Audio Graph Manager (AGM) Provides interfaces to allow TinyALSA-based mixer controls and PCM/compressed plug-ins to interact and enable audio features. AudioReach Graph Service (ARGS) Consists of the Graph Service Layer (GSL), Generic Packet Router (GPR), and acdb Management Layer (AML) modules. Handles initialization and creation of graphs, and creation of packets for sending a series of commands to the AudioReach Engine (ARE). Audio calibration database (acdb) Includes information about various audio use cases such as graphs, module calibration data, etc. The Apps processor parses acdb files to get the graph information used by ARE to enable the use case. AudioReach Engine (ARE) Modular framework that runs on the LPAI DSP. It helps set up, configure, and run signal processing modules for audio features. Qualcomm Audio Calibration Tool (QACT™ Platform) PC-based software that provides a GUI to visualize, configure, and store audio graphs in the acdb for audio use cases.

​
Next steps