Multi-Band Voice Compression to Preserve Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-frequency band wide dynamic range compression technologies in hearing aids often result in a decreased signal-to-noise ratio of output voice signals, leading to reduced recognition due to excessive loudness gain for noise signals.
Innovation Solution
A voice signal processing apparatus and method that filters input signals into different frequency bands, calculates and applies reference loudness gains based on wide dynamic range compression curves, and compares these gains with filtering loudness gains to select the smaller value for adjustment, thereby preventing signal-to-noise ratio drops and enhancing output signal recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-frequency band WDRC processing is applied to enhance loudness of output voice signal, then loudness gain for low loudness voice signals is improved, but signal-to-noise ratio drops due to excessive loudness gain for noise signals
Solution Approach 1:
The patent divides the voice signal into multiple frequency bands and processes each band independently with its own WDRC curve. By segmenting the signal processing, the system can apply different loudness gain characteristics to different frequency ranges, allowing enhancement of speech frequencies while controlling noise amplification in other bands.
Solution Approach 2:
The patent applies different WDRC curves tailored to specific frequency bands, where each curve is optimized for the characteristics of that band. This local optimization allows the system to enhance loudness where needed (speech frequencies) while maintaining or reducing gain where it would harm signal-to-noise ratio (noise-dominated frequencies).
2Measurement precision
If loudness gain is increased for low loudness voice signals to improve recognition, then loudness of output voice signal is enhanced, but loudness gain for noise signals also increases causing discomfort
Solution Approach 1:
The patent dynamically adjusts loudness gain based on the input signal characteristics in each frequency band. The WDRC processing continuously adapts the gain application, applying higher gain to low-loudness speech components while applying lower or negative gain to noise components, thereby improving recognition without causing discomfort from amplified noise.
Solution Approach 2:
The patent changes the loudness gain parameter selectively across different frequency bands and different signal levels. By modifying the gain parameter based on the specific frequency band and input loudness level, the system enhances speech recognition for low-loudness signals while preventing excessive amplification of noise signals that would cause discomfort.
Data Source
AI summary
A voice signal processing apparatus and a voice signal processing method are provided. A loudness of an input voice signal is detected to obtain a reference loudness. Reference loudness gains corresponding to frequency bands are calculated according to the reference loudness and wide dynamic range compression curves corresponding to the frequency bands. Loudnesses of filter signals of the frequency bands are adjusted according to the reference loudness gains of the frequency bands.