Audio Noise Estimation in the Log2 Domain for 16-Bit Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise estimation algorithms for audio signals require high computational complexity and dynamic range, especially when using fixed point processors, due to the need for precise data representation exceeding 32 bits, which is inefficient for battery-operated devices.
Innovation Solution
Converting energy values per band into the logarithmic domain, particularly the log2-domain, allows noise estimation using a minimum statistics algorithm with reduced precision, enabling efficient processing on fixed point processors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If noise estimation is performed using conventional algorithms on linear energy data, then measurement precision is improved, but device complexity increases due to requiring more than 32 bits dynamic range
Solution Approach 1:
The patent transforms the noise estimation problem from the linear domain to the logarithmic domain by applying a logarithmic transformation to the energy values. This parameter change allows the use of 16-bit fixed-point arithmetic instead of requiring more than 32 bits, significantly reducing processor complexity while maintaining adequate noise estimation precision for audio applications
2Measurement precision
If 40 bits dynamic range is used for noise estimation, then measurement precision is improved, but loss of energy increases due to higher computational complexity and storage demands
Solution Approach 1:
By changing the representation parameter from linear scale to logarithmic scale, the patent reduces the required dynamic range from 40 bits to 16 bits. This parameter transformation maintains sufficient precision for noise estimation while dramatically reducing computational complexity and energy consumption on battery-operated devices
Solution Approach 2:
The patent uses simplified 16-bit fixed-point arithmetic operations instead of complex high-precision calculations. This approach trades off some extreme precision for significantly lower computational cost and energy consumption, which is acceptable for noise estimation applications where perfect precision is not critical
3Device complexity
If 16 bit fixed point format is used for processing, then device complexity is reduced, but measurement precision deteriorates due to insufficient dynamic range
Solution Approach 1:
The patent applies a logarithmic transformation to compress the dynamic range of energy values into a form that can be accurately represented using 16-bit fixed-point format. This parameter change allows simple processors to handle noise estimation data with sufficient precision by operating in the logarithmic domain rather than the linear domain
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for estimating noise in an audio signal includes determining an energy value for the audio signal, converting the energy value into the log2-domain, and estimating a noise level for the audio signal based on the converted energy value directly in the log2-domain. The energy value is converted into the log2-domain as follows: En_log=⌊log21+En_lin⋅2N⌋2N └ x┘ floor (x), indicating the largest integer less than or equal to x, En_log energy value of band n in the log2-domain, En_lin energy value of band n in the linear domain, N quantization resolution. Determining the energy value includes obtaining a power spectrum of the audio signal by a combination of several transformations covering different parts of the spectrum.