Speech Signal Gain Adjustment for Silence-Area Noise Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech coding methods fail to effectively reduce noise artifacts in silence areas of speech signals, particularly when a voice activity detector (VAD) is not available, due to large quantization errors and limited accuracy from fixed-point implementations.
Innovation Solution
A method that involves obtaining incoming speech samples, smoothing the energy level, and adjusting the gain to reduce noise artifacts in silence areas by setting a gain value based on the spectral tilt parameter, with specific equations and conditions for determining silence and modifying the energy levels to minimize audible noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If speech compression is applied to reduce bit rate, then bandwidth is reduced, but speech quality is degraded
Solution Approach 1:
The patent applies different processing to different parts of the speech signal: silence regions are identified and treated differently from active speech regions. In silence regions, gain reduction is applied to suppress quantization noise, while in active speech regions, the original signal is preserved. This local differentiation allows quality improvement in critical regions without compromising overall speech quality.
Solution Approach 2:
The patent modifies the gain parameter specifically for silence regions based on spectral tilt characteristics. By changing the gain parameter locally in silence regions rather than uniformly across the entire signal, the system reduces quantization noise while maintaining speech quality in active regions, effectively resolving the bit rate vs. quality tradeoff.
2Quantity of substance
If quantization is applied to reduce data size, then bandwidth is reduced, but quantization errors increase noise in silence areas
Solution Approach 1:
The patent identifies that quantization noise is particularly problematic in silence regions where there is no speech signal to mask it. By detecting silence regions using spectral tilt analysis and applying gain reduction specifically to these regions, the patent converts the harmful quantization noise into acceptable background levels, effectively hiding the harm within the silence periods where it is least perceptible.
Solution Approach 2:
The patent performs preliminary detection of silence regions before the final signal reconstruction. By identifying silence regions in advance using spectral tilt analysis and pre-calculating appropriate gain values, the system can apply noise suppression in advance, preventing quantization noise from becoming audible rather than attempting to remove it after the fact.
3Device complexity
If fixed-point implementation is used to simplify processing, then device complexity is reduced, but measurement precision of silence areas deteriorates
Solution Approach 1:
The patent uses simple spectral tilt calculations and basic threshold comparisons to identify silence regions, rather than employing complex voice activity detection algorithms. This disposable, computationally inexpensive approach to silence detection is sufficient for the application's needs and maintains compatibility with fixed-point implementations while achieving the desired precision for noise suppression.
Data Source
AI summary
Methods and systems are presented for reducing effect of noise producing artifacts in a speech signal. An example method includes obtaining a plurality of incoming samples representative of a speech subframe, summing an energy level for each of the plurality of samples to generate a total input level, determining whether the speech subframe is in a silence area using the total input level, setting a gain value as a function of the total input level in response to the speech subframe being in the silence area, wherein the function results in the gain value increasing as the total input value increases and the gain value decreasing as the total input value decreases, multiplying the plurality of samples by the gain value to produce modified samples, generating a digital speech signal using the modified samples, and converting using a digital-to-analog converter, the digital speech signal into an analog signal.