Adaptive energy limiting for transient noise suppression

An apparatus comprising:

• a data interface to receive or transmit audio data over a data network;
• a limiter module to limit energy of audio signals that correspond to the audio data;
• a hardware-based processor associated with the data interface;
• and
• storage media storing processor-executable instructions that, responsive to execution by the hardware-based processor, implement an adaptive energy limiter configured to:
• set a ceiling level of the limiter module for the audio signals to a full-scale level;
• determine a maximum amplitude for a portion of a first audio signal, the first audio signal being one of the audio signals that pass through the limiter module;
• evaluate the portion of the first audio signal with a neural network to provide a voice likelihood estimate for the portion of the first audio signal;
• determine, based on the maximum amplitude and the voice likelihood estimate, that the portion of the first audio signal includes noise;
• decrease the ceiling level of the limiter module from the full-scale level to a decreased ceiling level in response to the determination that the portion of the first audio signal includes noise;
• and
• provide, to the limiter module, the decreased ceiling level to reduce the energy of the portion of the first audio signal.

A system comprising:

• a hardware-based processor operably associated with a data interface by which audio data is communicated;
• and
• storage media storing processor-executable instructions that, responsive to execution by the hardware-based processor, implement a conferencing service configured to:
• set a ceiling level of a limiter module through which audio signals of the audio data pass to a full-scale level;
• receive, via the data interface, a portion of the audio data that corresponds to a first audio signal, the first audio signal being one of the audio signals that pass through the limiter module;
• generate, based on the first audio signal, a frame of audio that corresponds to a duration of audio of the first audio signal;
• determine, for the frame of audio, a maximum amplitude of the first audio signal;
• determine, for the frame of audio, a voice likelihood estimate for the frame of audio using a neural network;
• determine, based on the maximum amplitude and the voice likelihood estimate, that the frame of audio includes noise;
• decrease the ceiling level of the limiter module from the full-scale level to a decreased ceiling level in response to the determination that the frame of audio includes noise;
• and
• provide, to the limiter module through which the audio signals pass, the decreased ceiling level to reduce an amount of energy of the first audio signal.