Machine-Learning AGC for Stable Speech Gain in Noisy Audio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern automatic gain control systems in conferencing technologies often fail to distinguish between speech and noise effectively, leading to undesirable gain changes that can dampen speech in noisy environments or amplify noise in quiet ones, resulting in poor audio quality and inefficient resource usage.
Innovation Solution
A machine learning-based system that analyzes audio data across multiple frequency ranges to differentiate between speech and noise, using a trained model to estimate the desired signal level and adjust gain values accordingly, thereby enhancing speech detection and maintaining a stable output volume.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional automatic gain control is used to adjust gain based on overall signal level, then output volume is stabilized, but speech clarity deteriorates in noisy environments because noise and speech are treated equally
Solution Approach 1:
The audio signal is segmented into multiple frequency bands using a filter bank, allowing separate analysis and processing of speech and noise components in different frequency ranges. This enables the system to identify speech-containing bands and apply gain control selectively, resolving the contradiction between overall volume stability and speech detection accuracy.
Solution Approach 2:
A machine learning model acts as an intermediary to classify frequency bands as speech or noise based on spectral features. This intermediary layer enables intelligent differentiation between speech and noise, allowing the AGC to preserve speech clarity while maintaining output volume stability through selective gain adjustment.
2Measurement precision
If gain is increased to amplify weak speech signals, then speech clarity improves, but background noise is also amplified resulting in poor audio quality
Solution Approach 1:
Different gain values are applied to different frequency bands based on their classification as speech or noise. Speech-containing bands receive higher gain to improve clarity, while noise-dominated bands receive lower gain or no amplification. This local differentiation resolves the contradiction between speech amplification and noise suppression.
3Object-generated harmful factors
If aggressive noise suppression is applied to remove background noise, then audio quality improves, but speech signal is also attenuated resulting in loss of speech content
Solution Approach 1:
The system dynamically adjusts gain values for different frequency bands based on real-time classification by the machine learning model. When speech is detected in a frequency band, the gain is increased to preserve or enhance speech content. When noise is detected, the gain is reduced to suppress noise. This dynamic adaptation resolves the contradiction between noise suppression and speech preservation.
Data Source
AI summary
Method includes receiving, through a plurality of channels, audio data corresponding to a plurality of frequency ranges; determining, for each channel's frequency ranges, speech audio and/or noise energy level using a model trained by machine learning; determining a speech signal with removed noise for each channel; determining one or more statistical values associated with an energy level of a channel's speech signal with the removed noise; determining a strongest channel that has highest statistical values associated with an energy level of a speech signal; determining that the one or more statistical values associated with the energy level of the strongest channel's speech signal satisfy a threshold condition; comparing statistical values associated with an energy level of a speech signal of each channel with those of the strongest channel; and determining whether to update a gain value for a channel based on the channel's statistical values associated with the energy level.