Signal normalization using loudness metadata for audio processing
One embodiment provides a method of signal normalization. The method comprises receiving an input content with a corresponding audio signal, and extracting loudness metadata from an audio signal corresponding to the input content. The method further comprises estimating, using a machine learning model, a peak-level amplitude based on the loudness metadata. The peak-level amplitude represents a maximum linear amplitude of the audio signal over an entire duration of the input content. The method further comprises determining a gain based at least on the peak-level amplitude, and applying the gain to the audio signal. The resulting gain-scaled audio signal is provided to one or more speakers coupled to or integrated in an electronic device for audio playback.
1 . A method of signal normalization, comprising:
receiving an input content with a corresponding audio signal;
extracting loudness metadata from an audio signal corresponding to the input content;
estimating, using a machine learning model, a peak-level amplitude based on the loudness metadata, wherein the peak-level amplitude represents a maximum linear amplitude of the audio signal over an entire duration of the input content;
determining a gain based at least on the peak-level amplitude; and
applying the gain to the audio signal, wherein the resulting gain-scaled audio signal is provided to one or more speakers coupled to or integrated in an electronic device for audio playback.
2 . The method of claim 1 , further comprising:
obtaining one or more desired user settings for the electronic device; and
obtaining one or more device settings representing a preset configuration of the electronic device.
3 . The method of claim 2 , wherein the gain is further based on at least one of the one or more desired user settings and the one or more device settings.
4 . The method of claim 1 , wherein the machine learning model is one of a least-squares optimal model, a linear regression model, a nonlinear regression model, or a fully-connected feed-forward neural network.
5 . The method of claim 1 , further comprising:
applying one or more audio post-processes to the gain-scaled audio signal before the audio playback.
6 . The method of claim 1 , wherein the one or more audio post-processes include at least one of perceptual bass enhancement, equalization, audio upmixing, spatial rendering, and digital-to-analog conversion.
7 . The method of claim 1 , wherein the peak-level amplitude is estimated without analyzing the entire duration of the input content.
8 . The method of claim 1 , wherein the loudness metadata is extracted from a header of the audio signal.
9 . A system of signal normalization, comprising:
at least one processor; and
a non-transitory processor-readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including:
receiving an input content with a corresponding audio signal;
extracting loudness metadata from an audio signal corresponding to the input content;
estimating, using a machine learning model, a peak-level amplitude based on the loudness metadata, wherein the peak-level amplitude represents a maximum linear amplitude of the audio signal over an entire duration of the input content;
determining a gain based at least on the peak-level amplitude; and
applying the gain to the audio signal, wherein the resulting gain-scaled audio signal is provided to one or more speakers coupled to or integrated in an electronic device for audio playback.
10 . The system of claim 9 , wherein the operations further include:
obtaining one or more desired user settings for the electronic device; and
obtaining one or more device settings representing a preset configuration of the electronic device.
11 . The system of claim 10 , wherein the gain is further based on at least one of the one or more desired user settings and the one or more device settings.
12 . The system of claim 9 , wherein the machine learning model is one of a least-squares optimal model, a linear regression model, a nonlinear regression model, or a fully-connected feed-forward neural network.
13 . The system of claim 9 , wherein the operations further include:
applying one or more audio post-processes to the gain-scaled audio signal before the audio playback.
14 . The system of claim 9 , wherein the one or more audio post-processes include at least one of perceptual bass enhancement, equalization, audio upmixing, spatial rendering, and digital-to-analog conversion.
15 . The system of claim 9 , wherein the peak-level amplitude is estimated without analyzing the entire duration of the input content.
16 . A non-transitory processor-readable medium that includes a program that when executed by a processor performs a method of signal normalization, the method comprising:
receiving an input content with a corresponding audio signal;
extracting loudness metadata from an audio signal corresponding to the input content;
estimating, using a machine learning model, a peak-level amplitude based on the loudness metadata, wherein the peak-level amplitude represents a maximum linear amplitude of the audio signal over an entire duration of the input content;
determining a gain based at least on the peak-level amplitude; and
applying the gain to the audio signal, wherein the resulting gain-scaled audio signal is provided to one or more speakers coupled to or integrated in an electronic device for audio playback.
17 . The non-transitory processor-readable medium of claim 16 , wherein the method further comprises:
obtaining one or more desired user settings for the electronic device; and
obtaining one or more device settings representing a preset configuration of the electronic device.
18 . The non-transitory processor-readable medium of claim 17 , wherein the gain is further based on at least one of the one or more desired user settings and the one or more device settings.
19 . The non-transitory processor-readable medium of claim 16 , wherein the machine learning model is one of a least-squares optimal model, a linear regression model, a nonlinear regression model, or a fully-connected feed-forward neural network.
20 . The non-transitory processor-readable medium of claim 16 , wherein the method further comprises:
applying one or more audio post-processes to the gain-scaled audio signal before the audio playback.