A variety of recent researches in Audio Emotion Recognition (AER) outlines high performance and retrieval accuracy results. However, in most works music is considered as the original sound content that conveys the identified emotions. One of the music characteristics that is found to represent a fundamental means for conveying emotions are the rhythm- related acoustic cues. Although music is an important aspect of everyday life, there are numerous non-linguistic and non- musical sounds surrounding humans, generally defined as sound events (SEs). Despite this enormous impact of SEs to humans, a scarcity of investigations regarding AER from SEs is observed. There are only a few recent investigations concerned with SEs and AER, presenting a semantic connection between the former and the listener’s triggered emotion. In this work we analytically investigate the connection of rhythm-related characteristics of a wide range of common SEs with the arousal of the listener using sound events with semantic content. To this aim, several feature evaluation and classification tasks are conducted using different ranking and classification algorithms. High accuracy results are obtained, demonstrating a significant relation of SEs rhythmic characteristics to the elicited arousal.
Sound Events and Emotions: Investigating the Relation of Rhythmic Characteristics and Arousal
1. Sound Events and Emotions: Investigating the
Relation of Rhythmic Characteristics and Arousal
K. Drossos 1 R. Kotsakis 2 G. Kalliris 2 A. Floros 1
1Digital Audio Processing & Applications Group, Audiovisual Signal Processing
Laboratory, Dept. of Audiovisual Arts, Ionian University, Corfu, Greece
2Laboratory of Electronic Media, Dept. of Journalism and Mass Communication, Aristotle
University of Thessaloniki, Thessaloniki, Greece
2. Agenda
1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
3. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
4. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Interaction with stimuli
Reactions
Emotions felt
5. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Interaction with stimuli
Reactions
Emotions felt
6. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Interaction with stimuli
Reactions
Emotions felt
7. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Interaction with stimuli
Reactions
Emotions felt
8. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Interaction with stimuli
Reactions
Emotions felt
9. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Structured form of sound
General sound
Interaction with stimuli
Reactions
Emotions felt
10. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Everyday life
Everyday life
Stimuli
Many stimuli, e.g.
Visual
Aural
Structured form of sound
General sound
Interaction with stimuli
Reactions
Emotions felt
11. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
12. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
13. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
14. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
15. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
16. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions
Music
Structured form of sound
Primary used to mimic and extent voice characteristics
Enhancing emotion(s) conveyance
Relation to emotions through (but not olny) MIR and MER
Music Information Retrieval
Music Emotion Recognition
17. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
18. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
19. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
20. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
21. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
22. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
Music and emotions (cont’d)
Research results & Applications
Accuracy results up to 85%
In large music data bases
Opposed to typical “Artist/Genre/Year” classification
Categorisation according to emotion
Retrieval based on emotion
Preliminary applications for synthesis of music based on emotion
23. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
24. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
25. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
26. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
27. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
28. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events
Sound events
Non structured/general form of sound
Additional information regarding:
Source attributes
Environment attributes
Sound producing mechanism attributes
Apparent almost any time, if not always, in everyday life
29. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
30. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
31. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
32. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
33. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
34. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound and emotions
From music to sound events (cont’d)
Sound events
Used in many applications:
Audio interactions/interfaces
Video games
Soundscapes
Artificial acoustic environments
Trigger reactions
Elicit emotions
35. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
36. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Objectives of current study
Data
Aural stimuli can elicit emotions
Music is a structured form of sound
Sound events are not
Music’s rhythm is arousing
Research question
Is sound events’ rhythm arousing too?
37. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Objectives of current study
Data
Aural stimuli can elicit emotions
Music is a structured form of sound
Sound events are not
Music’s rhythm is arousing
Research question
Is sound events’ rhythm arousing too?
38. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Objectives of current study
Data
Aural stimuli can elicit emotions
Music is a structured form of sound
Sound events are not
Music’s rhythm is arousing
Research question
Is sound events’ rhythm arousing too?
39. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Objectives of current study
Data
Aural stimuli can elicit emotions
Music is a structured form of sound
Sound events are not
Music’s rhythm is arousing
Research question
Is sound events’ rhythm arousing too?
40. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
41. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
42. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
43. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
44. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
45. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
46. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Experimental Procedure’s Layout
Layout
Emotional model selection
Annotated sound corpus collection
Processing of sounds
Pre-processing
Feature extraction
Machine learning tests
47. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
48. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
49. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
50. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
51. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
52. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Sound corpus & emotional model
General Characteristics
Sound Corpus
International Affective Digital Sounds (IADS)
167 sounds
Duration: 6 secs
Variable sampling frequency
Emotional model
Continuous model
Sounds annotated using Arousal-Valence-Dominance
Self Assessment Manikin (S.A.M.) used
53. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
54. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
55. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
56. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
57. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
58. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Pre-processing
General procedure
Normalisation
Segmentation / Windowing
Clustering in two classes
Class A: 24 members
Class B: 143 members
Segmentation/Windowing
Different window lengths
Ranging from 0.8 to 2.0 seconds, with 0.2 seconds increment
59. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Feature Extraction
Extracted features
Only rhythm related features
For each segment
Statistical measures
Total of 26 features’ set
60. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Feature Extraction
Extracted features
Only rhythm related features
For each segment
Statistical measures
Total of 26 features’ set
61. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Feature Extraction
Extracted features
Only rhythm related features
For each segment
Statistical measures
Total of 26 features’ set
62. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Feature Extraction
Extracted features
Only rhythm related features
For each segment
Statistical measures
Total of 26 features’ set
63. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Processing of Sounds
Feature Extraction
Extracted features
Only rhythm
related features
For each
segment
Statistical
measures
Extracted Features Statistical Measures
Beat spectrum Mean
Onsets Standard deviation
Tempo Gradient
Fluctuation Kurtosis
Event density Skewness
Pulse clarity
64. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Feature Evaluation
Objectives
Most valuable features for arousal detection
Dependencies between selected features and different window
lengths
Algorithms used
InfoGainAttributeEval
SVMAttributeEval
WEKA software
65. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Feature Evaluation
Objectives
Most valuable features for arousal detection
Dependencies between selected features and different window
lengths
Algorithms used
InfoGainAttributeEval
SVMAttributeEval
WEKA software
66. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Feature Evaluation
Objectives
Most valuable features for arousal detection
Dependencies between selected features and different window
lengths
Algorithms used
InfoGainAttributeEval
SVMAttributeEval
WEKA software
67. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Classification
Objectives
Arousal classification based on rhythm
Dependencies between different window lengths and
classification results
Algorithms used
Artificial neural networks
Logistic regression
K Nearest Neighbors
WEKA software
68. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Classification
Objectives
Arousal classification based on rhythm
Dependencies between different window lengths and
classification results
Algorithms used
Artificial neural networks
Logistic regression
K Nearest Neighbors
WEKA software
69. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Machine learning tests
Classification
Objectives
Arousal classification based on rhythm
Dependencies between different window lengths and
classification results
Algorithms used
Artificial neural networks
Logistic regression
K Nearest Neighbors
WEKA software
70. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
71. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Feature Evaluation
General results
Two groups of features formed, 13 features each group
An upper and a lower group
Inner group rank not always the same
Upper and lower groups constant for all window lengths
72. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Feature Evaluation
General results
Two groups of features formed, 13 features each group
An upper and a lower group
Inner group rank not always the same
Upper and lower groups constant for all window lengths
73. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Feature Evaluation
General results
Two groups of features formed, 13 features each group
An upper and a lower group
Inner group rank not always the same
Upper and lower groups constant for all window lengths
74. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Feature Evaluation
General results
Two groups of features formed, 13 features each group
An upper and a lower group
Inner group rank not always the same
Upper and lower groups constant for all window lengths
75. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Feature Evaluation
Upper group
Beatspectrum std
Event density std
Onsets gradient
Fluctuation kurtosis
Beatspectrum gradient
Pulse clarity std
Fluctuation mean
Fluctuation std
Fluctuation skewness
Onsets skewness
Pulse clarity kurtosis
Event density kurtosis
Onsets mean
76. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Lowest accuracy score: 71.26%
77. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Lowest accuracy score: 71.26%
78. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Lowest accuracy score: 71.26%
79. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Window length: 1.0 second
Algorithm utilised: Logistic regression
Lowest accuracy score: 71.26%
80. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Window length: 1.0 second
Algorithm utilised: Logistic regression
Lowest accuracy score: 71.26%
81. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Classification
Classification
General results
Relative un-corelation of features
Variations regarding different window lengths
Accuracy results
Highest accuracy score: 88.37%
Window length: 1.0 second
Algorithm utilised: Logistic regression
Lowest accuracy score: 71.26%
Window length: 1.4 seconds
Algorithm utilised: Artificial Neural network
82. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
83. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
84. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
85. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
86. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
87. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
88. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature evaluation
Features
Most informative features:
Rhythm’s periodicity at auditory channel (fluctuation)
Onsets
Event density
Beatspecturm
Pulse clarity
Independent from semantic content
93. 1. Introduction
Everyday life
Sound and emotions
2. Objectives
3. Experimental Sequence
Experimental Procedure’s Layout
Sound corpus & emotional model
Processing of Sounds
Machine learning tests
4. Results
Feature evaluation
Classification
5. Discussion & Conclusions
6. Feature work
94. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature work
Features & Dimensions
Other features related to arousal
Connection of features with valence
95. Introduction Objectives Experimental Sequence Results Discussion & Conclusions Feature work
Feature work
Features & Dimensions
Other features related to arousal
Connection of features with valence