Introduction to Microprocesso programming and interfacing.pptx
A Musical System For Emotional Expression
1. A Musical System For Emotional Expression
António Pedro Oliveira
University of Coimbra, Portugal
2. 2
Outline of the Presentation
Introduction
Background
Emotion-Driven Music Engine (EDME)
Conclusion
3. 3
Outline of the Presentation
Introduction
Background
Emotion-Driven Music Engine (EDME)
Conclusion
4. Motivation
Emotions are widely accepted as being an
important factor in the society
Music is almost everywhere and it is a powerful
stimulus capable of influencing our emotions
Computational systems with the capability of
producing music with an appropriate emotional
content have an enormous application potential
4
Introduction Background EDME Conclusion
5. Aim
Conceive a computational system for the
control of the emotional content of
produced music, so that it expresses a given
emotional specification
Music: solely instrumental
5
Introduction Background EDME Conclusion
6. 6
Outline of the Presentation
Introduction
Background
Emotion-Driven Music Engine (EDME)
Conclusion
7. Emotional Expression with
Music
Four approaches:
Music Transformation
Music Composition
Music Selection/Classification
Hybrid Approaches
Our approach:
Hybrid that consists in combining
selection/classification with transformation
7
Introduction Background EDME Conclusion
8. 8
Outline of the Presentation
Introduction
Background
Emotion-Driven Music Engine (EDME)
Conclusion
12. Initial phase
Manually Built Knowledge Base (short
version)
Happy music: high loudness, major scale
Sad music: violin, slow tempo
Activating music: high loudness, fast tempo
Relaxing music: low loudness, slow tempo
12
Introduction Background EDME Conclusion
14. Experiments
First Experiment - Preliminary Evaluation of
the Classification Module
Hypothesis: There is a small amount of features
that may predict arousal/valence
Valence: 2 features, CC – 0.76
Arousal: 4 features, CC – 0.77
14
Introduction Background EDME Conclusion
15. Experiments
Second experiment (two parts)
First part - Extended Evaluation of the Classification
Module
Hypothesis: There is a small amount of features that
may predict arousal/valence
Valence: 4 features, CC – 0.70
Arousal: 3 features, CC – 0.77
15
Introduction Background EDME Conclusion
16. Experiments
Second part - Analysis of Audio Features
Hypothesis: There are audio features emotionally-
relevant
Valence: Spectral Sharpness and Loudness are important
Arousal: Spectral Similarity, Spectral Dissonance and
Spectral Sharpness are important
16
Introduction Background EDME Conclusion
17. Experiments
Third experiment (three parts)
First part - Improvement of the Classification Module
Hypothesis: There is a small amount of features that
may predict arousal/valence
Valence: 5 features, CC – 0.69
Arousal: 3 features, CC – 0.71
17
Introduction Background EDME Conclusion
18. Experiments
Second part – Evaluation of the Transformation
Algorithms
Hypothesis: It is possible to change musical features to
transform emotional content
High positive coefficients for tempo were confirmed
The increase of register correlates positively with
valence and negatively with arousal
Some of the features can be helpful in finding scales more
appropriate to some emotions
Instruments are essentially relevant to the arousal
Change from normal to staccato articulation has a
correlation with the increase of valence
18
Introduction Background EDME Conclusion
19. Experiments
Third part – Melodic Analysis
Hypothesis: The analysis of the melodic line alone turns
the emotionally-relevant features more visible
19
Introduction Background EDME Conclusion
Emotional Dimension CC Features
Valence – data of first experiment 0.79 5 features
Valence – data of second
experiment
0.62 5 features
Valence – data of third experiment 0.41 4 features
Arousal – data of first experiment 0.85 4 features
Arousal – data of second
experiment
0.72 4 features
Arousal – data of third experiment 0.54 3 features
20. Calibration and Validation
Hypothesis: 13 features are enough to
discriminate emotional expression
Valence: 7 features, CC – 0.85
Arousal: 6 features, CC – 0.83
Use SAM to obtain emotional answers
Controlled environment
Statistical analysis:
System’s classification and subject’s classification
are probably measuring the same concept
20
Introduction Background EDME Conclusion
21. 21
Outline of the Presentation
Introduction
Background
Emotion-Driven Music Engine (EDME)
Conclusion
22. Contributions
The system proposed has the advantage of
being able to produce outputs of acceptable
quality quite independently from the music
base
It is also quite flexible: the music base can be
completely redefined to adapt to the specific
needs of a given scenario
22
Introduction Background EDME Conclusion
23. Contributions
The system is also reliable, thanks to the
experimental calibration using different
subjects
We adopted both the discrete and
dimensional representation of emotions
We used techniques of human emotional
recognition for validation and calibration of
the system
23
Introduction Background EDME Conclusion
24. Future Work
Emotion-Driven Music Composition
Artificial Intelligence approaches
Test in applications contexts
Healthcare
Entertainment
24
Introduction Background EDME Conclusion
Nowadays, music is almost everywhere, and the most interesting fact is that it is a powerful stimulus capable of influencing our emotions
These systems are usable in every context where there is a need to create environments capable of inducing certain emotional experiences
Conceive a computational system for the control of the emotional content of produced music, so that it expresses a given emotional specification
Music: solely instrumental MIDI
Produced music is solely instrumental, which has been proved to be sufficient to express desired emotions (Kimura, 2002)
Emotions: Valence and arousal
There are four different principal approaches used in the generation of music with appropriate emotional content
Our approach is hybrid that consists in combining classification/selection with transformation
Classification of pre-composed music and subsequent selection
The quality of the answers is dependent on the original music base
The picture has a simple representation of how the classification and subsequent selection work
We have a cluster of musical pieces represented in a bi-dimensional space
In the case of our work this is an emotional space
Our approach selects musical pieces closer to the desired emotion
Three modules (segmentation, feature extraction, classification); three auxiliary structures (pre-composed music, music base, knowledge base)
Pre-composed music is obtained from the Internet
It is:
1) Segmented
2) Classified with the help of the knowledge base
3) Stored in a music base
This stage had four modules and four auxiliary structures
Software uses desired emotion to:
Select music that expresses emotion closer to the desired one
Transform and sequence selected segments to form musical structures
Musical structures are synthesized with the help of a library of sounds and played to the listener
we had an initial phase that consisted in building manually a first version of the knowledge base by considering empirical data collected from works of Music Psychology
the set of features was selected manually, according to what we learned from the literature about their relative importance to emotional expression.
We also defined weights for the features in accordance with the literature.
Four types of music
Five types of musical features
First two types of music has to do with the valence dimension of emotions
Second two types of music has to do with the arousal dimension of emotions
Initial phase and experiments were made with the objective of making the knowledge base
We carried out three experiments conducted via Web
In a non-controlled environment
With subjects from different areas
They were made in order to build regression models and to successively refine their set of features and corresponding weights
CC is measure of the linear correlation (dependence) between two variables X and Y, giving a value between +1 and −1
30 musical pieces
30 subjects
it seems that the 13 considered features can discriminate well both valence and arousal of each music
we can infer that the ex- periments conducted via online have a high degree of reliability, despite the fact of being made in a non-controlled context.
The statistical results using Kappa and Cramer’s V do not only confirm the reliability of this calibration/validation study but also the reliability of the experiments conducted with the help of online questionnaires
Contexts of application:
Theatre, films, video-games and healthcare contexts.
Contexts of application:
Theatre, films, video-games and healthcare contexts.