Method, system, and medium for affective music recommendation and composition

1 . A processor-executed method for generating an audio stream for inducing an affective state change in a listener, comprising:

identifying a current affective state of the listener;

identifying a target affective state of the listener;

identifying an affective trajectory from the current affective state to the target affective state;

using a trained segment identification machine learning model to identify a first audio segment likely to induce in the listener a desired affective response corresponding to at least an initial portion of the affective trajectory when the first audio segment is presented to the listener as an auditory stimulus;

using a trained affect inference machine learning model to infer an inferred new affective state based on the current affective state and a set of audio feature values of the first audio segment;

identifying an updated affective trajectory from the inferred new affective state to the target affective state;

using the trained segment identification machine learning model to identify a subsequent audio segment likely to induce in the listener a subsequent desired affective response corresponding to at least an initial portion of the updated affective trajectory when the subsequent audio segment is presented to the listener as an auditory stimulus;

generating the audio stream based at least in part on the first audio segment and the subsequent audio segment;

sending audio stream data based on the audio stream to a listener device; and

causing the audio stream to be presented to the listener via at least one speaker of the listener device.

2 . The method of claim 1 , wherein:

the trained affect inference machine learning model is trained using training data comprising:

training audio feature data corresponding to a plurality of training audio segments, and

affective state data gathered from one or more human subjects in association with exposure of each human subject to each of a plurality of audio stimuli corresponding to the plurality of training audio segments.

3 . The method of claim 2 , wherein:

the one or more human subjects comprises the listener.

4 . The method of claim 3 , further comprising, after sending the audio stream data to the listener device:

receiving updated current affective state data from the listener; and

training the trained affect inference machine learning model using runtime training data comprising:

audio feature data corresponding to each of the first audio segment and the subsequent audio segment; and

the updated current affective state data.

5 . The method of claim 4 , wherein:

the audio feature data comprises music information retrieval (MIR) data.

6 . The method of claim 2 , wherein:

the trained segment identification machine learning model comprises a reinforcement learning model.

7 . The method of claim 2 , wherein:

the trained segment identification machine learning model comprises a deep learning neural network.

8 . The method of claim 2 , wherein:

the training audio feature data comprises music information retrieval (MIR) data.

9 . The method of claim 1 , wherein:

identifying the listener's target affective state comprises:

receiving target affective state data from the listener via the listener device; and

identifying the listener's target affective state based on the target affective state data.

10 . A non-transitory processor-readable medium containing instructions for executing the method of claim 1 .

11 . A non-transitory storage medium containing the audio stream generated by the method of claim 1 .

12 . The method of claim 1 , wherein:

the trained segment identification machine learning model is trained using reward data received from the trained affect inference machine learning model; and

the trained affect inference machine learning model generates the reward data by:

inferring an inferred affective response of the listener to a set of audio feature values of the audio stream; and

generating the reward data based on a comparison of the inferred affective response to the desired affective response.

13 . The method of claim 1 , wherein:

the audio stream data comprises recommendation data recommending the audio stream.

14 . The method of claim 1 , wherein:

the audio stream data comprises the audio stream.

15 . The method of claim 1 , wherein:

identifying the listener's current affective state comprises:

receiving affective self-evaluation data from the listener via the listener device; and

identifying the listener's current affective state based on the affective self-evaluation data.

16 . The method of claim 1 , wherein:

identifying the listener's current affective state comprises:

receiving physiological data correlated with one or more physiological states of the listener; and

identifying the listener's current affective state based on the physiological data.

17 . A system comprising:

a processor system; and

a memory system having stored thereon instructions executable by the processor system to cause the system to perform the method of claim 1 .

18 . The method of claim 1 , wherein:

the set of audio feature values of the first audio segment comprises music information retrieval (MIR) data.