System and method for detecting stress in audio data

A computerized system for audio stress detection, the system comprising:

• a memory;
• and a computer processor configured to:
• calculate a buffer length based on one or more audio attributes;
• extract an audio buffer from an audio data item based on the calculated buffer length;
• calculate a dynamic buffer length using one or more overlapping frames within the audio buffer, wherein the processor is to:
• clean the audio buffer, the cleaning comprising removing one or more audio parts based on a signal-to-noise threshold;
• divide the cleaned buffer into the one or more overlapping frames;
• compare the dynamic buffer length to a default frame length;
• and
• adjust the dynamic buffer length based on the comparison and based on one or more of the overlapping frames;
• extract an updated buffer from the audio buffer based on the dynamic buffer length;
• and
• predict, by a machine learning model, a stress level for the audio buffer based on the updated buffer.

A method for audio stress detection, the method comprising:

• in a computerized system comprising a memory and a processor:
• computing, by the processor, a segment size based on one or more audio characteristics;
• extracting, by the processor, an audio segment from an audio file based on the calculated segment size;
• computing, by the processor, a dynamic audio segment size using one or more overlapping frames within the audio segment, comprising:
• cleaning, by the processor, the audio segment, the cleaning comprising filtering one or more audio parts based on a signal-to-noise threshold;
• dividing, by the processor, the cleaned segment into the one or more overlapping frames;
• comparing, by the processor, the dynamic segment size to a default frame size;
• and
• adjusting, by the processor, the dynamic segment size based on the comparison and based on one or more of the overlapping frames;
• extracting, by the processor, an updated segment from the audio segment based on the dynamic segment size;
• and
• predicting, by a machine learning model, a stress score for the audio segment based on the updated segment.