You are here: UK Essays » Essays » Music » Emotions Classical Western

Cookie Information

Privacy Information

Music Essays

Emotions Classical Western

Abstract

It is well known that music carries emotions and the emotions felt are common to most of the people listening to it. However, they do vary over the geographical region from where the music comes and the people listening to it. Music from around the world brings varied emotions and is perceived differently by different sets of people. When the same scale or set of tunes are played in a certain way defined by the Indian classical music is different than the same played in western classical style. The people would perceive both kinds of music with difference. This makes detection of emotions in music challenging. A system that recognizes the emotions will also have to recognize the type of music being played.

Emotion recognition in music can be achieved by detecting the traits followed by music with similar emotions in them. These properties of music can be the properties of the sound: pitch, amplitude, timbre; rhythm, and frequency intervals of selected notes; and the way a tune is played: staccato, vibrato, and slide. Thus the style of playing can determine the emotions generated by the music. This project is based on the interval detection of the played notes to determine the emotion in musical pieces.

The scope of this project has been limited to Indian Classical Music for building a statistical notion of the intervals used to create specific emotions in music. A raga in Indian classical music is an equivalent of a scale in Western music. All ragas are associated with some kind of emotion. Understanding the pattern of the intervals of those scales and building a statistical notion of emotions based on these intervals will help understand the emotions of the ragas. Such statistics will show the interval usage of a particular emotion. The various emotions that are associated will be calmness, aggressiveness, seriousness, loneliness, etc. Most of the time the emotion perceived by humans falls between calm and aggressive emotions. Aggressive is often used for energetic feeling. Such statistics can now be applied to various music styles around the world. Assuming that most emotions perceived are more or less similar, this model will be able to detect emotions from all kinds of music. To make the recognition more accurate, one can find the error vectors for individual music type and then apply a model after error correction using the error vectors. Error vectors can be generated by taking the difference of perceived statistics and system generated statistics.

Thus from the intervals used for a specific type of emotion, style of playing and some amplitude analysis can bring out the emotion hidden behind the musical piece. It is not necessary that a musical piece would have only one emotion and so the level of different emotions must be calculated by analyzing short time intervals. Introduction

Human beings use language as a tool to communicate. The language does not always have a literal dictionary meaning. It becomes meaningful when certain emotions are associated with it. Recognition of this hidden abstract property is important for a lot of applications in the real world. Depending on the senses excited, emotions can be classified as visual, tangible, and auditory. The emotions felt through hearing can have two sources: speech and music. Recognition of emotions in human speech has been studied over a long period from and engineering point of view; however, emotion recognition in music has not been explored as extensively. Over centuries, musicians have studied emotions in music and practiced them in performances from a completely different perspective. This paper discusses an approach to automatic emotion recognition in music. The base of this theory uses the knowledge of ‘Raga’ and its associated emotions from Indian Classical music.

The Indian Classical music is one of the most ancient classical music in the world. It has morphed and evolved over centuries. Since most of the work associated here is from the northern Indian classical music, it will be henceforth referred to as Hindustani music. Hindustani music is highly structured so as to classify groups of notes based on melodies, associated emotions and time of performance. There are ‘Thath’ which are only group of notes from which one can derive different Raga. Some rules dictate the structure of these Thath and Raga as well. A Raga is a group of 5 or more different notes, equivalent to a scale in Western Classical music. However, there is more to the Raga than just the notes. A direct translation of one of the descriptions is “That particularity of notes and melodic movements, or that distinction of melodic sound by which one is delighted, is Raga “[1]. Thus, a Raga is a set of notes with a particular melodic movement to give a characteristic emotion that touches the mind and pleases the ear.

A Raga is more precise in its definition but gives enough flexibility to the musician for improvisation and composition. All these hundreds of possible Raga have a certain emotion associated with them. These emotions or notions of feelings – certain ‘mood’ – come from daily life. Just as the sweetness of sugar, sugar cane or jaggery cannot be separately described, but experienced; the detailed emotions in Raga can only be experienced. Broadly these basic emotions can be classified as happy, sad, aggressive and calm. Ragas have been classified on the basis of the notes they contain. These have also been give attributes of the basic emotions. Hence, the emotions can be associated with the note content of a Raga. In Hindustani music, it is mentioned that a particular note generates emotion with the help of other note. This small detail gives an idea of an interval. Since all Raga use a particular note (Do) as their basis, individual notes are equivalent to their intervals.

Hindustani music uses a different notation for representing the notes. They are as follows:

The first row is the note name, equivalent name in western music and generally associated frequency notes. Hindustani music is composed mainly in three octaves. The lower octave is called ‘Mandra’, middle ‘Madhya’ and upper ‘Taar’. The Sa note of the middle octave if at ‘C’ will have frequency 261.63 Hz. The basic 7 notes are {Sa, Re, Ga, Ma, Pa, Dha, Ni}. The _k means ‘Komal’ which is flat note and _t is ‘Tivra’ which is sharp note. This convention will be used in the paper and in the supporting program as well. Please see appendix for detailed frequencies.

Hindustani music associates some emotions to the Raga based on the notes they have. For example, a Raga with Re_k and Dha_k will have serious and sad emotions associated with them; that with Ma_t is joyous; Ga_k and Ni_k is sad and aggressive [4]. By doing a statistical analysis of emotions involved in these Raga, one can build a recognition system that analyses the current intervals in the music and match then with those found in the ‘Raga’. It will be safe to assume that the emotion associated with that particular ‘Raga’ will be that in the music being played. The emotion in a music piece is not always unique but can be a mixture and so we can detect the % of each emotion in music piece from this database by match.

The emotion largely depends on the perception of an individual. It varies over the geographical region from where the music comes and the people listening to it. Music from around the world brings varied emotions and is perceived differently by different sets of people. Assumption here is that this perceived emotion is different in detail but similar in broader sense. Hence, this model should recognize the right emotion from a non Hindustani musical piece. The next section discusses how a musical piece from a file is processed so as to extract intervals and compare for recognition.

The recorded music file can be a .wav or .au file which can be read in MATLAB. Preferably a noise free source is welcome. It is acceptable to have the sound of interest with enough amplitude over noise. Emotions can many times be dictated by amplitude, vibrato, slide, or energy in sound. For simplicity of the problem, they have not been considered in this model.

Short Term Segmenting: This is a moving window over which a Fourier transform will be performed. The design of this window has to change according to the sampling frequency of the music file. The size of this window directly relates to the proper detection of the notes with respect to the speed at which they are played. Smaller the size better the resolution, but higher the number of calculations.

Spectrum Analysis: This module uses Fourier transform and calculates the FFT values for select frequencies where the notes are located. Hence, the frequency spectrum is discrete and hence allows good recognition of the note. Spectrum can be used to see the flow of notes over period of time. It gives the analyzed freq vector for a particular time frame.

Note Detection: Note detection from the spectrum vector is done using the max amplitude criteria. Since there are no harmonics or chords in consideration, so as to keep the project simple, only one note can be excited at a moment. There can be a period of silence between notes. Only the max amplitude note is retained and rest information discarded.

Remove temporal factor: Since a normal music is assumed, there will not be long periods where only one note is active. Moreover the change in the notes brings out music. Hence, the temporal factor is removed buy taking a mode over the set of time frames. The mode is used over small sets so as to detect the fast moving notes. However, due to select frequency FFT in spectrum analysis, valid frequency values are evaluated even during silence periods. This will create error in interval calculation. Hence a limiting factor has to be put to see if the mode has enough high frequency for it being a valid played note.

Interval Analysis: The core part of the simulation is to determine the emotion. Here all the changes in notes are calculated and a percentage relation is built. This percentage is then compared to the trained data and then shown as output as a visual emotion meter. The training data is obtained from analyzing and hand calculating the intervals involved in many Raga. All their contained intervals and the frequency of those intervals are near matched to currently playing piece segment and analyzed for emotion. So first a database is created and then these training weights are defined.

Results

The sad Raga ShivRanjani is detected as sad, Bhupali which has all kinds of emotions shows all emotions as well. The graph shows emotion bar of each music file. A Greek[2] and an Arabic[3] music pieces were tried to see the results.

Raga ShivRanjani

Raga Bhupali

Conclusions

The program has shown crude matching results for a music from other regions. There is more to emotions in music than only intervals. This method is still in early stages of development. It requires more research. A music from one region of the world can be similar to another, however it is still crude. Finer adjustments can be made by improving the detection of different traits. Some experiments were done by using stoccato and slide, however they need some more work for recognition. Considering the amplitude and energy at a given time will also give better analysis for the emotion separation between aggressive and calm natures.

References

• THE RAGA GUIDE – JOEP BOR

• http://members.fortunecity.com/ananast0/midi/page4.html - for greek music

• http://cyberegypt.com/music.htm - for Arabic music

• RAGA DARSHAN – SANGEET PUSTAK

• http://www.phy.mtu.edu/~suits/notefreqs.html - for frequency table

Sign up and be the first to receive our latest offers: