Speech Signal Processing

 

Introduction

Speech Signal Processing (SSP) Engine is essential in voice communication and speech recognition applications.

When a voice is captured on a microphone, it is mixed in with various noise signals from the surrounding environment.

As a result, the quality of the audio data is poor. Because poor-quality voices cannot be used in speech recognition programs, the SSP engine improves the quality of audio data by removing the noise signals.

 

Process of SSP Engine

 

SSP Engine by LG AI Platform supports the following features.

Feature Description
Multi-channel PCM data support Supports single- or multi-channel microphones. It can be applied to a variety of devices using single- or dual-channel PCM data files as reference signals.
Optimal algorithm for speech recognition pre-processing The SSP engine is equipped with high-performance Acoustic Echo Cancellation (AEC) and Noise Reduction (NR) algorithms, and is suitable for pre-processing speech recognition.
Excellent AEC performance AEC effectively removes the echo from various forms of speaker output.
Excellent NR performance NR is equipped with both beamforming (BF) and source separation (SS) techniques, guaranteeing the best performance even in a multi-channel microphone environment.
Flexible gain control Gain can be adjusted through various techniques such as Auto Gain Control, Dynamic Level Control, Limiter, and Parametric EQ to improve the reliability of speech recognition.
Distant speech recognition Recognizes small and distant voices.
Noise and echo reduction Effectively reduces noise and echo, even in noisy environments. Noise/echo reduction rate and voice retention rate can be adjusted according to your preferences.
Multiple operation modes Keyword, Recognition, and Beamforming modes are provided to be made applicable to various devices with a speech recognition feature.
WNR (Wind Noise Reduction) support Selectively removes wind noise due to road surfaces, as well as any sudden noises due to other non-linear environments.
UI-based tuning tool Tune and simulate the parameters of each module using the UI-based SSP Engine Tuning Tool.

 

Engine Structure

SSP Engine consists of SSP Core and Config Data. It receives PCM data as input values ​​and generates PCM output data with improved sound quality.

 

Architecture of SSP Engine

 

Examples of Use

SSP Engine is used in a variety of areas that require voice-based services.

 

  • Improve voice quality for speech recognition of a robot platform

​Improves the quality of speech recognition by removing background noise.

 

SSP Engine usecase

  • Improves voice quality for speech recognition of home appliances

Improves the quality of speech recognition by removing background noise from the surrounding environment.

Improve voice quality for speech recognition of appliances