Speech Bandwidth Extension Using Voiced-Unvoiced Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The transition from narrowband to wideband speech in VoIP networks is hindered by the need for efficient speech quality enhancement during an intermediate coexistence period, where existing signal processing models consume high processing power with limited performance improvement.
Innovation Solution
A speech bandwidth extension system and method that applies algorithms to narrowband signals to extend high and low frequencies, using pre-processing, signal classification, adaptive signal extension, and post-processing to generate wideband speech signals, leveraging sigmoid functions and Automatic Gain Control for optimal harmonics generation and energy control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If CELP speech coding extension models are used to improve speech quality during narrowband/wideband coexistence, then speech quality is improved, but processing power consumption increases significantly
Solution Approach 1:
The patent transforms the narrowband speech signal parameters (sampling frequency 8kHz, bandwidth 200-3400Hz) into wideband parameters (sampling frequency 16kHz, bandwidth 50-7500Hz) through mathematical transformation and spectral extension, achieving quality improvement without requiring full wideband encoding processing power
Solution Approach 2:
The patent creates an artificial copy of the missing high-frequency spectrum by extrapolating and transforming the existing narrowband spectral characteristics, generating synthetic wideband components that mimic natural speech frequencies without capturing actual wideband signal data
2Manufacturing precision
If existing signal processing models are used to extend speech bandwidth, then speech quality is improved, but device complexity increases
Solution Approach 1:
The patent divides the bandwidth extension process into distinct stages: pre-processing (spectral analysis), signal classification (voiced/unvoiced detection), adaptive signal extension (frequency-specific processing), and post-processing (spectral smoothing), allowing each segment to be optimized independently with reduced overall complexity
Solution Approach 2:
The patent employs dynamic adaptive processing where the extension algorithm adjusts its parameters based on real-time signal characteristics, using voiced/unvoiced classification to selectively apply different extension strategies, thereby reducing average processing complexity compared to static full-bandwidth extension
Data Source
AI summary
There is provided a method or a device for extending a bandwidth of a first band speech signal to generate a second band speech signal wider than the first band speech signal and including the first band speech signal. The method comprises receiving a segment of the first band speech signal having a low cut off frequency and a high cut off frequency; determining the high cut off frequency of the segment; determining whether the segment is voiced or unvoiced; if the segment is voiced, applying a first bandwidth extension function to the segment to generate a first bandwidth extension in high frequencies; if the segment is unvoiced, applying a second bandwidth extension function to the segment to generate a second bandwidth extension in the high frequencies; using the first bandwidth extension and the second bandwidth extension to extend the first band speech signal beyond the high cut off frequency.