Neural Network Quantization for Audio Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise suppression methods face challenges in accurately and efficiently estimating noise, particularly in real-time audio streaming, leading to increased latency and battery consumption.
Innovation Solution
The development of a system and method that trains and optimizes a neural network model to quickly process noisy audio, using indirection buffers to avoid costly concatenation processes, and iteratively improves the model through retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex noise suppression techniques are used to improve noise estimation accuracy, then noise suppression quality is improved, but processing time and power consumption increase
Solution Approach 1:
The patent applies parameter changes by quantizing the neural network model weights to lower precision formats (e.g., from 32-bit floating point to 16-bit or 8-bit integers). This reduces the computational complexity and power consumption of the model while maintaining acceptable noise suppression performance, directly resolving the contradiction between accuracy and energy consumption.
Solution Approach 2:
The patent creates a quantized copy of the neural network model that uses simplified data representations. This quantized model processes audio data with reduced computational requirements compared to the full-precision model, enabling real-time processing on resource-constrained devices while maintaining functional accuracy for noise suppression.
2Measurement precision
If complex noise suppression techniques are used to improve noise estimation accuracy, then noise suppression quality is improved, but processing time increases causing latency
Solution Approach 1:
By changing the precision parameters of the neural network weights through quantization, the patent reduces the number of computational operations required per audio frame. This enables faster processing of audio data in real-time, reducing latency while maintaining sufficient accuracy for effective noise suppression in streaming applications.
Solution Approach 2:
The patent segments the audio processing into discrete frames and processes them independently through the quantized neural network. This frame-based segmentation allows for efficient real-time processing where each frame can be handled quickly with reduced computational complexity, preventing cumulative latency buildup.
3Measurement precision
If larger neural network models are used to improve noise suppression quality, then accuracy is improved, but device complexity and processing requirements increase
Solution Approach 1:
The patent changes the parameter representation from high-precision floating point numbers to lower-precision formats through quantization. This reduces the memory footprint and computational complexity of the neural network model, allowing deployment on devices with limited resources while preserving the essential noise suppression functionality.
Solution Approach 2:
The patent uses quantized model weights that require less computational resources to process each audio frame. This approach prioritizes efficiency over raw model capacity, using a simplified computational representation that achieves acceptable performance with significantly reduced device complexity and processing requirements.
Data Source
AI summary
The disclosed computer-implemented method may include capturing, by a computing device, a media clip. The method may also include dividing, by the computing device, the media clip into a set of frames, wherein each frame may include an audio portion of the media clip of a predetermined length of time. Additionally, the method may include performing, by the computing device, a noise suppression process on each frame of the set of frames using a trained neural network model, wherein the trained neural network model is quantized to use input tensors. Finally, the method may include creating, by the computing device, a clean media clip based on the noise suppression process. Various other methods, systems, and computer-readable media are also disclosed.