Dynamic sound adjustment based on noise floor estimate

A system for estimating a steady-state noise floor in a signal, the system comprising:

• a steady-state noise estimator comprising one or more processing devices, the steady-state noise estimator configured to:
• receive a plurality of representations of the signal corresponding to samples of the signal within a frame of predetermined time duration,
• estimate a power spectral density (PSD) for each of a plurality of frequency bins, wherein the PSD for a particular frequency bin is estimated based on a smoothing parameter calculated from a noise estimate for the particular frequency bin as obtained from samples corresponding to a preceding frame,
• generate, based on the PSD for each of the plurality of frequency bins, an estimate of the steady-state noise floor;
• and
• a spectral flatness estimator configured to compute a measure of spectral flatness associated with the samples within the frame, the measure of flatness being calculated based on PSDs calculated for at least a portion of the plurality of frequency bins,
• wherein the steady state noise estimator is further configured to:
• determine, based on feedback from the spectral flatness estimator, if the measure of spectral flatness satisfies a threshold condition, wherein the threshold condition is selected to emphasize steady-state noise across the portion of the plurality of frequency bins over spectral peaks in particular frequency bins in the same portion,
• responsive to determining that the measure of spectral flatness satisfies the threshold condition, compute an updated estimate of the steady-state noise floor, and
• responsive to determining that the measure of spectral flatness does not satisfy the threshold condition, maintaining the steady-state noise floor estimate as obtained from the samples corresponding to the preceding frame.

One or more non-transitory machine-readable storage devices having encoded thereon computer readable instructions for causing one or more processing devices to perform operations comprising:

• receiving a plurality of representations of a signal corresponding to samples of the signal within a frame of predetermined time duration;
• estimating a power spectral density (PSD) for each of a plurality of frequency bins, wherein the PSD for a particular frequency bin is estimated based on a smoothing parameter calculated from a noise estimate for the particular frequency bin as obtained from samples corresponding to a preceding frame;
• generating, based on the PSD for each of the plurality of frequency bins, an estimate of a steady-state noise floor;
• computing a measure of spectral flatness associated with the samples within the frame, the measure of spectral flatness being calculated based on PSDs calculated for at least a portion of the plurality of frequency bins;
• and
• determining if the measure of spectral flatness satisfies a threshold condition, wherein the threshold condition is selected to emphasize steady-state noise across the portion of the plurality of frequency bins over spectral peaks in particular frequency bins in the same portion;
• responsive to determining that the measure of spectral flatness satisfies the threshold condition, computing an updated estimate of the steady-state noise floor;
• and
• responsive to determining that the measure of spectral flatness does not satisfy the threshold condition, maintaining the steady-state noise floor estimate as obtained from the samples corresponding to the preceding frame.