Methods and apparatus for audio equalization
View Patent ↗Methods, apparatus, systems and articles of manufacture are disclosed for audio equalization. Example apparatus disclosed herein include a volume adjuster to apply equalization adjustments to an audio signal to generate an equalized audio signal, the equalization adjustments output from a neural network in response to an input feature set; a thresholding controller to: detect an irregularity in a frequency representation of the audio signal after application of the equalization adjustments, the irregularity corresponding to a change in volume between adjacent frequency values exceeding a threshold; and adjust a volume at a first frequency value of the adjacent frequency values to reduce the irregularity; an equalization (EQ) curve generator to generate an EQ curve to apply to the audio signal when the irregularity has been reduced; and a frequency to time domain converter to output the equalized audio signal in a time domain based on the EQ curve.
1. An apparatus comprising:
a volume adjuster to apply a plurality of equalization adjustments to an audio signal to generate an equalized audio signal, the plurality of equalization adjustments output from a neural network in response to an input feature set including an average volume representation of the audio signal;
a thresholding controller to:
detect an irregularity in a frequency representation of the audio signal after application of the plurality of equalization adjustments, the irregularity corresponding to a change in volume between adjacent frequency values exceeding a threshold; and
adjust a volume at a first frequency value of the adjacent frequency values to reduce the irregularity;
an equalization (EQ) curve generator to generate an EQ curve to apply to the audio signal when the irregularity has been reduced; and
a frequency to time domain converter to output the equalized audio signal in a time domain based on the EQ curve.
2. The apparatus of claim 1 , further including an energy calculator to:
determine a first root mean square (RMS) value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments;
determine a second RMS value of the frequency representation of the audio signal after reduction of the irregularity; and
determine a ratio between the second RMS value and the first RMS value.
3. The apparatus of claim 1 , further including a volume normalizer to:
determine whether a ratio between (1) a first RMS value of the frequency representation of the audio signal after reduction of the irregularity and (2) a second RMS value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments exceeds a threshold associated with acceptable change in energy of the audio signal; and
in response to the ratio exceeding the threshold, apply a normalization of a gain of the frequency representation of the audio signal.
4. The apparatus of claim 1 , wherein the plurality of equalization adjustments includes a plurality of volume adjustment values corresponding to a plurality of frequency ranges.
5. The apparatus of claim 1 , wherein the thresholding controller is to:
select a plurality of frequency values in the frequency representation of the audio signal;
determine a plurality of volume values associated with the plurality of frequency values;
determine a second derivative of the volume over the plurality of frequency values; and
in response to an absolute value of the second derivative exceeding the threshold, adjust the volume at the first frequency value of the adjacent frequency values to reduce the irregularity.
6. The apparatus of claim 1 , wherein the plurality of equalization adjustments is based on at least reference audio signals, EQ curves, and tags associated with a plurality of audio engineers that generated the EQ curves, and wherein the neural network determines the plurality of equalization adjustments based on an inference associated with at least the reference audio signals, the EQ curves, and the tags associated with the plurality of audio engineers.
7. The apparatus of claim 6 , wherein the input feature set includes the average volume representation of the audio signal and average standard deviation measurements for frequency bins of the frequency representation of the audio signal.
8. A non-transitory compute readable storage medium comprising instructions which, when executed, cause one or more processors to at least:
apply a plurality of equalization adjustments to an audio signal to generate an equalized audio signal, the plurality of equalization adjustments output from a neural network in response to an input feature set including an average volume representation of the audio signal;
detect an irregularity in a frequency representation of the audio signal after application of the plurality of equalization adjustments, the irregularity corresponding to a change in volume between adjacent frequency values exceeding a threshold;
adjust a volume at a first frequency value of the adjacent frequency values to reduce the irregularity;
generate an equalization (EQ) curve to apply to the audio signal when the irregularity has been reduced; and
output the equalized audio signal in a time domain based on the EQ curve.
9. The non-transitory computer readable storage medium of claim 8 , wherein the instructions, when executed, cause the one or more processors to:
determine a first root mean square (RMS) value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments;
determine a second RMS value of the frequency representation of the audio signal after reduction of the irregularity; and
determine a ratio between the second RMS value and the first RMS value.
10. The non-transitory computer readable storage medium of claim 8 , wherein the instructions, when executed, cause the one or more processors to:
determine whether a ratio between (1) a first RMS value of the frequency representation of the audio signal after reduction of the irregularity and (2) a second RMS value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments exceeds a threshold associated with acceptable change in energy of the audio signal; and
in response to the ratio exceeding the threshold, apply a normalization of a gain of the frequency representation of the audio signal.
11. The non-transitory computer readable storage medium of claim 8 , wherein the plurality of equalization adjustments includes a plurality of volume adjustment values corresponding to a plurality of frequency ranges.
12. The non-transitory computer readable storage medium of claim 8 , wherein the instructions, when executed, cause the one or more processors to:
select a plurality of frequency values in the frequency representation of the audio signal;
determine a plurality of volume values associated with the plurality of frequency values;
determine a second derivative of the volume over the plurality of frequency values; and
in response to an absolute value of the second derivative exceeding the threshold, adjust the volume at the first frequency value of the adjacent frequency values to reduce the irregularity.
13. The non-transitory computer readable storage medium of claim 8 , wherein the plurality of equalization adjustments is based on at least reference audio signals, EQ curves, and tags associated with a plurality of audio engineers that generated the EQ curves, and wherein the neural network determines the plurality of equalization adjustments based on an inference associated with at least the reference audio signals, the EQ curves, and the tags associated with the plurality of audio engineers.
14. The non-transitory computer readable storage medium of claim 13 , wherein the input feature set includes the average volume representation of the audio signal and average standard deviation measurements for frequency bins of the frequency representation of the audio signal.
15. A method comprising:
applying a plurality of equalization adjustments to an audio signal to generate an equalized audio signal, the plurality of equalization adjustments output from a neural network in response to an input feature set including an average volume representation of the audio signal;
detecting an irregularity in a frequency representation of the audio signal after application of the plurality of equalization adjustments, the irregularity corresponding to a change in volume between adjacent frequency values exceeding a threshold;
adjusting a volume at a first frequency value of the adjacent frequency values to reduce the irregularity;
generating an equalization (EQ) curve to apply to the audio signal when the irregularity has been reduced; and
outputting the equalized audio signal in a time domain based on the EQ curve.
16. The method of claim 15 , further including:
determining a first root mean square (RMS) value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments;
determining a second RMS value of the frequency representation of the audio signal after reduction of the irregularity; and
determining a ratio between the second RMS value and the first RMS value.
17. The method of claim 15 , further including:
determining whether a ratio between (1) a first RMS value of the frequency representation of the audio signal after reduction of the irregularity and (2) a second RMS value of the frequency representation of the audio signal prior to application of the plurality of equalization adjustments exceeds a threshold associated with acceptable change in energy of the audio signal; and
in response to the ratio exceeding the threshold, applying a normalization of a gain of the frequency representation of the audio signal.
18. The method of claim 15 , wherein the plurality of equalization adjustments includes a plurality of volume adjustment values corresponding to a plurality of frequency ranges.
19. The method of claim 15 , further including:
selecting a plurality of frequency values in the frequency representation of the audio signal;
determining a plurality of volume values associated with the plurality of frequency values;
determining a second derivative of the volume over the plurality of frequency values; and
in response to an absolute value of the second derivative exceeding the threshold, adjusting the volume at the first frequency value of the adjacent frequency values to reduce the irregularity.
20. The method of claim 15 , wherein the plurality of equalization adjustments is based on at least reference audio signals, EQ curves, and tags associated with a plurality of audio engineers that generated the EQ curves, and wherein the neural network determines the plurality of equalization adjustments based on an inference associated with at least the reference audio signals, the EQ curves, and the tags associated with the plurality of audio engineers.