AI-powered speech noise suppression! Dramatically improving speech recognition rate

Narrow down by specifying conditions

現在2149件がヒットしています。check

Basic Infineon

Introduction

In recent years, devices and applications using voice recognition have expanded rapidly. However, in real-world environments, there is a lot of ambient noise and reverberation, making speech noise suppression a bottleneck in recognition accuracy.
In the past, the most common method was to generate an inverse phase noise to cancel out the audio signal, but this method had the problem that it was only effective when the desired audio signal was known.

Infineon's PSOC™ Edge solves this problem using AI (machine learning).
By using a learning model to automatically identify and separate "necessary speech" from "noise," it achieves highly accurate noise suppression even in situations that were previously difficult to achieve.
In addition, AI models can be provided as samples, allowing for easy evaluation and implementation in a development environment using ModusToolbox®.

This makes it possible to develop a highly accurate voice noise suppression system in a shorter time than before, and dramatically improves voice recognition rates.

What is an Audio Front End?

Audio Front End (AFE) is a general term for a group of software components for audio signal processing provided by Infineon.
In voice input devices and voice recognition applications, it performs preprocessing to make the voice picked up from the microphone clearer and easier to understand.

Infineon offers the Audio Front-End Configurator as a dedicated tool to easily configure and optimize these AFE components.
The tool's GUI allows you to visually combine individual processing blocks and adjust parameters to design the optimal audio processing pipeline.
When combined with PSOC™ Edge, noise suppression and beamforming can be achieved using edge AI.

Main functional blocks

The Audio Front End allows you to freely design your configuration by enabling/disabling the following functional blocks:

Functional Blocks	Overview
Equalizer	Adjusts the frequency response to emphasize or attenuate specific frequency bands.
Acoustic Echo Canceller (AEC)	This function detects the spontaneous sound generated by the speaker and eliminates echoes, which is effective for hands-free calling.
Beamforming	It emphasizes the sound coming from a specified direction and suppresses the sound coming from other directions. It utilizes multiple microphone inputs.
Dereverb	It reduces the reverberation that occurs in rooms, etc., and extracts clear audio.
Noise Suppression	Using AI model calculations, ambient noise is reduced in real time.

By freely combining these blocks, you can build a flexible audio processing pipeline that suits your application needs.
In addition, the parameters for each block can be intuitively set on the GUI, making it possible to customize audio processing without specialized DSP knowledge.

Audio Front End Overview

Audio Front End construction tools

The Audio Front End construction tool is Infineon's development tool It can be used from ModusToolbox™. The Audio Front-End Configurator GUI allows you to configure each functional block.
An example of what can be done with the GUI is
- Enabling/disabling function blocks
- Various parameter settings
is possible.

Sample image: Audio Front-End Configurator GUI

Left: Parameter and function block settings Right: Beamforming settings

Listen to the effect of noise suppression in the demo audio

You can hear the effect of noise suppression by PSOC™ Edge through actual audio.
The same audio source is compared "before processing" and "after noise removal by AI."
We recommend using earphones or headphones to experience the difference in sound.

Audio before noise removal (audio + noise)

Audio after noise removal (AI processed)

After AI processing, background noise is significantly reduced, making the speaker's voice clearer.
This technology is effective in improving speech recognition accuracy and in a variety of applications.

Inquiry

If you are interested in this kit, the development tool ModusToolbox™, or the Audio Front-End Configurator, please contact us using the form below.

Site Search