NS (Noise Suppression)

Supported ICs

Overview

Noise suppression is one of the AFE’s features, to suppress ambient noise, It can improve speech recognition performance in noisy environments and optimizes signal quality for communication systems.

Algorithm Description

Modes

The NS in AIVoice consists of two mode: signal processing and neural network methods, can be configured through the parameter ns_mode:

  • Signal processing mode: Based on traditional signal processing techniques, optimized for stationary noise (e.g., fan noise, white noise). Default mode.

  • Neural network mode: Use deep learning models to additionally suppress non-stationary noise (e.g., impacts, transient noises). Only supports Voice communication mode

Enable Guidelines

  • Speech recognition

    • Disabled by default as speech distortion may degrade wake-up/recognition accuracy.

    • When the environment noise is high or the device generates large stationary noise, the NS module is advised to enabled.

  • Voice communication

    • Except for special requirements, the NS module needs to be enabled for call clarity.

  • Spectrum when NS module is off and on:

    ../../../_images/afe_ns_spectrogram.png

Configurations

The complete AFE configuration file is located at ${aivoice_lib_dir}/include/aivoice_afe_config.h. The NS-related configurations include:

afe_mode_e afe_mode;                    // AFE mode, for ASR or voice communication.
bool enable_ns;                         // NS(Noise Suppression) module switch

// NS module parameter
afe_ns_mode_e ns_mode;                  // NS mode, signal process or NN method. NN method is only supports voice communication mode.
afe_ns_cost_mode_e ns_cost_mode;        // low cost mode means 1channel NR and poorer noise reduction effect
afe_ns_aggressive_mode_e ns_aggressive_mode;        // higher mode means more stationary noise suppression but more distortion