Acquiring speech features for predicting emotional severity of adverse events on individuals

A method, comprising:

• receiving, by at least one computing device, an audio sample that includes an utterance of a user;
• performing, by the at least one computing device, a feature extraction on the audio sample to extract a plurality of acoustic emotion features using a base model, the base model using a first convolution neutral network (CNN) to extract an initial set of emotion features from the audio sample;
• generating, by the at least one computing device, a plurality of emotion level predictions for an emotion type based at least in part on the plurality of acoustic emotion features being provided to an emotion specific model, the emotion specific model comprising a second CNN and being trained to identify a respective level of the emotion type;
• and
• determining, by the at least one computing device, an emotion classification for the audio sample based at least in part on the plurality of emotion level predictions, the emotion classification comprises the emotion type and a level associated with the emotion type.

A system, comprising:

• at least one computing device;
• and
• machine-readable instructions stored in the memory that, when executed, cause the at least one computing device to at least:
• receive an audio sample that includes an utterance of a user, the audio sample comprises a label, the label comprising an emotion level for a particular emotion type;
• perform a feature extraction on the audio sample to extract a plurality of acoustic emotion features using a base model, the base model using a first convolution neutral network (CNN) to extract an initial set of emotion features from the audio sample;
• generate a plurality of emotion level predictions for an emotion type based at least in part on the plurality of acoustic emotion features being provided to an emotion specific model;
• determine an emotion classification for the audio sample based at least in part on the plurality of emotion level predictions, the emotion classification comprises the emotion type and a predicted level associated with the emotion type;
• and
• update a parameter associated with the emotion specific model based at least in part on a comparison between the emotion classification and the label of the audio sample.