From Telemann’s compositional ideas to performance quali-ties and audio footprint signal processing analysis. The paradigm of Telemann’s Prelude from Fantasia Nº 2 à Travers sans Basse

Das ideias de composição de Telemann às qualificações de desempenho e análise de processamento de sinais de sinais de áudio. O paradigma do Prelúdio de Telemann de Fantasia Nº 2 à Travers sans Basse

Sofia Balula Dias
University of Lisbon
Interdisciplinary Centre for the study of human Performance (CIPER)
Estrada da Costa 1499-002 Cruz Quebrada/Dafundo, Portugal
Leontios J. Hadjileontiadis
Aristotle University of Thessaloniki
Faculty of Technology
Signal Processing and Biomedical Technology Unit
Thessaloniki Greece


FMH – ULisbon - University of Lisbon
AUT - Aristotle University of Thessaloniki

Retirado de:

RESUME: In this work, a bilateral approach is applied to Telemann’s music, combining information both from the music analysis domain and transformation techniques of the sound signal. The Prelude from Telemann’s Fantasia No 2 à Travers sans Basse is used as the paradigm basis and its characteristics that relate with the melodic and harmonic projection (music analysis axis) are seen through the perspectives of dynamics, pitch, timbre, similarity, novelty, chroma and predicted emotional impact, and the way they affect the micro-macrostructure of the piece, resulting in the final sound output (advanced signal processing axis). Two recordings from eminent flutists [Kuijken (flute traverso); Rampal (modern flute)] provided the audio data. The proposed approach could serve as an alternative handling of the sound ontologies, aiming at the optimum arrangement and realization of the structural intentions of a composer, as reflected in the interpretation of the performer and mediated to the audience.  

KEYWORDS: Telemann’s Fantasia No 2 à Travers sans Basse; melodic and harmonic projection; advanced signal processing; novelty; predicted emotional impact.  

RESUMO: Neste trabalho, é aplicada à música de Telemann uma abordagem bilateral, que combina informações do domínio de análise musical e das técnicas de transformação do sinal de som. O Prelúdio da Fantasia No 2 de Telemann é usado como paradigma de base, e as características que estão relacionadas com a projeção melódica e harmónica (eixo de análise musical) são exploradas ao nível das dinâmicas, frequência, timbre, similaridade, “novidade”, croma e impacto emocional previsto, e a forma como estes afetam a micro-macrosestrutura da peça, resultando na saída do som final (eixo de processamento de sinal avançado). Foram usadas como dados de áudio duas gravações de virtuosos flautistas [Kuijken (traverso); Rampal (flauta moderna)]. Esta proposta pode servir como uma abordagem alternativa às ontologias do som, visando uma disposição ideal e concretização das intenções estruturais de um compositor, refletido na interpretação do executante e mediado para o público.

PALAVRAS-CHAVE: Fantasia No 2 de Telemann para Traverso sem Baixo; projeção melódica e harmónica; processamento de sinal avançado; “novidade”; impacto emocional previsto.

1. Introduction

Georg Philipp Telemann (14 March 1681–25 June 1767) is one of the most prolific composers in history (at least in terms of surviving oeuvre) and was considered by his contemporaries to be one of the leading German composers of the time. He was a flautist himself and one of the most important composers for this instrument. Moreover, he was the only composer of his time to write fugues and a French overture for solo flute, and his set of 12 Fantasias for solo flute [12 Fantaisies à Travers, sans Basse, TWV 40:2-13, 1727/8-publ. 1732/33 (Brown, n.d.; Kuyjken, 1987; Zohn, 2008)] was written specifically for the flute, leaving a clear footprint in the repertoire, similar to the one of J. S. Bach’s “Well Tempered Clavier” for keyboard instruments. Despite the believes by the 18th century composers about the flute incapability to perform alone due to its inability to create and sustain harmony (Brown, n.d), Telemann explored the potential of this instrument, setting the Fantasias as one of the most representative work in the repertoire for solo flute from the 18th century.

Telemann, being a flutist himself, incorporated in his Fantasias power elements of Rhetoric of his era, by effectively using the affective potentiality of each key within the construction of the whole set, producing a desired effect to the sonic space. Moreover, tempo alterations add to this, since a fast movement is followed by a slower one, and vice-versa. In fact, by writing each Fantasia in a different tonality (ascending from A major through A minor, B minor, Bb major, C major, D minor, D major, E minor, E major, F# minor and G major to G minor), he applies a cyclic, almost encyclopedic presentation of the key sequence, so to increase the emotional variance across the whole set. As Brown notes:


Contrasting moods of joy, brilliance, fun, passion, pride, seriousness, coldness, plaintiveness, serenity, tenderness, delicacy and charm, as well as rustic and courtly dancing are reflected in not only the choice and juxtaposition of keys, but also the chosen register and motives, such as bold arpeggios, characterful rhythms or more lyrical legato lines. (Brown, n.d., p. 2)

The 12 Fantasias have an important pedagogical importance, as are suitable both to beginners as well as to professional flutists. What it is important to both skill levels, however, is the structural information that Telemann encrypts behind his music score, which facilitates the comprehension of his intention and guides the performance of the whole piece. In this vein, this work aims at identifying the flow from Telemann’s compositional ideas to performance qualities and auditory footprint. It uses, as a working paradigm, the Prelude (first movement) from the Fantasia No 2 à Travers sans Basse to exemplify the potentiality behind the proposed approach. Moreover, two performances of the Prelude are analyzed, one using flute traverso (performed by Barthold Kuijken) and another that uses modern flute (performed by Jean-Pierre Rampal). Advanced signal processing techniques are applied to the selected audio recordings, in order to provide a tangible space for the compositional and performance qualities to be further clarified. The latter, pave the way for further deepening in the music of Telemann and, in more general, to the Baroque music.


2. The research focus

The research axes of this paper are depicted in Fig. 1. The main research question relates with the information flow that starts from the composer’s site, who expresses his intentions about the final music output creating the music score. However, as expected, what is notated on the score has to be processed by the performer’s interpretation and expression, embedded in the performance, delivering the music (as a final audio stimulus) to the audience. Since Telemann’s music is written for flute traverso, apart from the differences in the idiosyncrasy, skills and style of performance, the core element of the shifting from the flute traverso to the modern flute sets another facet of analysis in this work. Finally, since the audio output encapsulates the compositional and performance characteristics across time, a dynamic analysis is performed across the structural elements of the Prelude, unfolding micro-, meso- and macro-resolution characteristics along with predicted emotional impact.


Fig. 1 – The research axes of the proposed work.


3. Methodology

To implement the proposed research path of Fig. 1, the following methodological axes were adopted:


Axis 1: Music analysis of the Prelude including description of:

·       the implicit harmonies and compound lines; and

·       the harmonic and melodic environment.

Axis 2: Advanced signal processing analysis of the audio signals from the two performances (i.e., flute traverso and modern flute) including:

·       dynamics analysis (envelope structure);

·       timbral analysis (attack, slope, brightness);

·       pitch analysis (main pitch distribution across time);

·       similarity analysis (similarity in spectral characteristics);

·       novelty analysis (temporal succession of moments, each characterized by particular musical properties);

·       chromagram analysis (self-organizing map projection of chromagram); and

·       emotional analysis (activity, valence, and tension space).

For the Axis 1, the music analysis of Prelude by Da Silva (2012) is adopted and followed as a basis to build upon the interpretation of Axis 2 analysis results. With regard to the latter, the analysis was based upon the following approaches:


Dynamics analysis: From an audio waveform the envelope can be computed, which shows the global outer shape of the signal. It is particularly useful in order to show the long-term evolution of the signal, expressing the change of dynamics across the time duration. The signal is segmented at 1-second successive windows and the normalized across segments envelope is used here, i.e., with respect to the global maxima across all the segments of the audio signal. A full-wave rectification, reflecting all the negative lobes of the signal into the positive domain, followed by a low-pass auto-regressive filter and a down-sampling with a rate of 16 provides a smooth estimation of the signal envelope.


Timbral analysis: This relates with the attack characteristics, i.e., attack time and the attack slope (as a ratio between the magnitude difference at the beginning and the ending of the attack period, and the corresponding time difference), as defined in Fig. 2. Moreover, the brightness, i.e., the amount of spectral energy above a specific cut-off frequency  (Juslin, 2000) is also estimated.


Fig. 2 – The definition of the (a) attack time and (b) attack slope.


Pitch analysis: The pitch is estimated via the autocorrelation function of the audio signal across the segments, keeping the most prominent peaks at each segment.

Similarity analysis: The structural features within a piece of music have been visualized with a self-similarity matrix (Foote et al., 2002), a matrix that shows the degree of similarity between different parts of a musical piece. Let denote a vector representing any musical feature at instant . The self-similarity matrix     is defined as

where  denotes any similarity measure. By definition, the matrix is symmetrical across its diagonal. Figure 8(a) illustrates schematically the calculation of a self-similarity matrix. Usually, a distance measure is adopted as a similarity measure of the representation vectors. The later could be derived from spectral-based parameterizations, linear prediction, Mel-Frequency Cepstral Coefficient (MFCC) [1] or psychoacoustic considerations. A simple measure is the Euclidian distance in the parameter space. In addition, useful metric is the scalar product of the vectors (internal product). This will be large if the vectors are both large and similarly oriented. To remove the dependence on magnitude (and hence energy, given our features), the product can be normalized to give the cosine of the angle between the parameter vectors. The cosine measure ensures that windows with low energy, such as those containing silence, can still yield a large similarity score, which is generally desirable and is given bellow:



Fig. 3 – (a) Estimation of the Self-Similarity Matrix, and (b) the Gaussian checkerboard kernel used in novelty.


The contents of a self-similarity matrix can be visualized as a square using different colors to indicate different degrees of similarity. The resulting matrix is visualized so that: bright shades of gray stand for high degrees of similarity and dark shades for low degrees of similarity. Regions of high self-similarity appear as bright squares on the diagonal. Repeated sections will be visible as bright off-diagonal rectangles. If the work has a high degree of repetition, this will be visible as diagonal stripes or checkerboards, offset from the main diagonal by the repetition time. The diagonal line at indicates that each frame is maximally similar to itself. Any possible periodicities visible in the spectrogram can also be visible in the similarity matrix.
Novelty Curve: Convolution along the main diagonal of the similarity matrix using a Gaussian checkerboard kernel (Fig. 8(b)) yields a novelty curve that indicates the temporal locations of significant textural changes. The result is a one-dimensional function of time (frame index). Intuitively, the correlation emphasizes regions with strong self-similarity while penalizing regions with significant cross-similarity. Peak detection applied to the novelty curve returns the temporal position of feature discontinuities that can be used for the actual segmentation of the audio sequence.
Chromagram analysis: The chromagram, also called Harmonic Pitch Class Profile, shows the distribution of energy along the pitches or pitch classes. Projection of the normalized chromagram into a self-organizing map [2], trained with the Krumhansl-Kessler profiles (modified for chromagrams) (Toiviainen and Krumhansl, 2003), provides a pseudo-color map, where colors correspond to Pearson correlation values for the corresponding pitches.
Emotional analysis: For the emotion analysis, a multi-dimensional space is adopted where the three dimensions are activity (or energetic arousal, which is characterized by vigor and energy), valence (a pleasure-displeasure continuum) and tension (or tense arousal, which is characterized by tension and nervousness) (Eerola et al., 2009). An attempt to predict such description of emotion, based on the analysis of the audio and musical contents of the recordings is realized here. In fact, following a systematic statistical methodology, a mapping has been proposed by Eerola et al. (2009) between each dimension and various sets of audio and musical features. In particular, the five factors contributing to the activity score are: Root Mean Square (RMS) averaged along frames; Maximum value of summarized fluctuation; Spectral centroid averaged along frames; Spectral spread averaged along frames; Entropy of the smoothed and collapsed spectrogram, averaged along frames. The five factors contributing to the valence score are: Standard deviation of RMS along frames; Maximum value of summarized fluctuation; Key clarity averaged along frames; Mode averaged along frames; Averaged spectral novelty. The five factors contributing to the tension score are: Standard deviation of RMS along frames; Maximum value of summarized fluctuation; Key clarity averaged along frames; Averaged Harmonic Change Detection Function (HCDF), i.e., the flux of the tonal centroid; Averaged novelty from unwrapped chromagram.


4. Data characteristics and implementation issues

The two audio recordings used as the data sources here were extracted as .m4a files (@44.1kHz) from the corresponding publicly available YouTube video links. In particular, the link was used for the Barthold Kuijken’s performance, whereas the link: for the Jean-Pierre Rampal’s one. These two performers were selected as very well known worldwide ones for their skills in flute performance and their special interpretation of flute traverso and modern flute, respectively. The advanced signal processing analysis was carried out using Matlab 2016a (The Mathworks, Inc., Natick, USA), and especially the MIRtoolbox I.3 (for emotion analysis only) and MIRtoolbox I.7 (for the rest of the analysis) ( The cutoff frequency  for the calculation of the brightness was selected equal to 1200Hz, according to the spectral characteristics of the two recordings.


5. Results

5.1 Axis 1: Music analysis results

Fantasia No 2, following the form of a Sonata da Chiesa (Porter et al., 2008), is a four-movement piece, i.e., Prelude (slow), Fugue (fast), Adagio (German style fully ornamented, slow), and Allegro (France Dance – Bourée, fast). The first movement plays the role of the Prelude to the Fugue that follows, constituting a separate movement, following Bach’s later style of Preludes and Fugues.

The Prelude examined here, is cast in 3/4 meter, and it has a serious and majestic character, adopting a slow harmonic rhythm, with basically one chord per measure (Da Silva, 2012), and a contrapuntal technique of compound lines, as shown in the reductive analysis of Fig. 4 (adopted from (Da Silva, 2012)).


Fig. 4 – Reductive analysis of the Prelude (Da Silva, 2012).


Moreover, as Da Silva (2012) notes, Telemann plays with suspensions and sequence (i – iv9-8 in A minor, and I – IV9-8 in G major) in the first two measures, as shown in Fig. 5.


Fig. 5 – Implied suspensions in the first bars of the Prelude (Da Silva, 2012).


Another important harmonic aspect is the evolution of the sequences that follow, in which voice leading is alternated with changes, not only in the rhythmical figuration and texture, but also in subverting the rules of counterpoint, trying to increase harmonic surprises and dramatic changes of texture (Schulenberg, 2008). A such example is given in Fig. 6.


Fig. 6 – Ascending motion with the bass (D – E – F), in bars 6-8 of the Prelude (Da Silva, 2012).


The basic chord progression (i – iv7-6 – V/III – III7-6 – VI – iv – i64 – iv – V) leads from i in A minor, to V, which is actually being resolved to the tonic i at the second movement; hence, fulfilling the role of a prelude, so to prepare the second movement and the fugue that follows.


5.2 Axis 2: Advanced signal processing analysis results

Duration and dynamics analysis results: Figures 7(a) and (b) depict the normalized envelope of the Prelude audio signals from Kuijken and Rampal, respectively. As it is clear from the time duration of each signal, Kuijken’s interpretation lasts 45s, whereas Rampal’s 35s; hence, Rampal plays the piece 28.5% faster than Kuijken. Moreover, looking at the Kuijken’s interpretation (Fig. 7(a)) the maximum lies at 32.4s whereas in Rampal’s one it lies at the 24.3s, quite close to each other, taking into account the duration analogy (24.3s×1.285=31.2s). It is interesting to notice that the position of the climax by Telemann is towards the golden ratio , so to maximize the psychological impact to the audience. From the two interpretations, Rampal’s is closer to  with a ratio of   and with Kuijken’s exhibiting .


Fig. 7 –The normalized envelope of the Prelude audio signals from (a) Kuijken and (b) Rampal.

Moreover, looking at the distribution of dynamics across the piece it is clear that it differs between the two interpretations. Apparently, Kuijken sustains the dynamic at levels lower than 0.8 so to leave ample space for the climax to appear (Fig. 7(a)), whereas Rampal uses quite high dynamic even from the beginning of the piece (peaks at 3.2s and 6.6s with envelope value of 0.96), reduces the dynamics around 16.5s, in order to prepare the climax (Fig. 7(b)). Furthermore, what follows the maximum peaks differs, as Kuijken sound decays faster towards the pause (bar 10 in Fig. 4), whereas Rampal creates a fluctuation in his decay with peaks preceding maximum at 24.9s (0.86), 25.4s (0.62) and 26.3s (0.36). These duration and dynamics differences show alternative ways of interpretation of Telemann’s intension to construct the Prelude. Perhaps Kuijken follows a more introvert sound handling, without covering the whole dynamic range of the soundscape, considering the Prelude as the initial sound gesture that gradually leads to the next movement (Fugue). On the other hand, Rampal shows a more extrovert approach, considering the Prelude as an autonomous sound entity and evokes more “internal episodes” that gradually build the music language and sustain the development potential for the movements that follow.

Timbral analysis: Figures 8(a) and (b) depict the attack time of the onsets (Fig. 2(a)) of the Prelude audio signals from Kuijken and Rampal, respectively. Clearly, both performers have similar attack times (mostly lower than 0.2s), whereas Kuijken exhibits a double maximum attack time (1.2s) compared to Rampal (0.6s), used when entering after the pause of bar 10 in Fig. 4. Rampal uses a higher attack time in the preparation of the climax and sustains quite low attack time for entering after the pause of bar 10 in Fig. 4. These differences show that, unlike Rampal, Kuijken prolongs, in a way, the impact of the pause after the climax, so the listener could gradually shift from the silence to the sound again, mostly echoing the psychological impact of the sound energy potential that was just released, when rapidly shifting from the maximum dynamic to silence.   


Fig. 8 – The attack time of the Prelude audio signals from (a) Kuijken and (b) Rampal.


Figures 9(a) and (b) depict the attack slope of the onsets (Fig. 2(b)) of the Prelude audio signals from Kuijken and Rampal, respectively. From the latter, it is clear that Rampal (on average) uses higher slopes than Kuijken, clearly showing the use of a different embouchure style (perhaps supported by the difference in the two instruments), in order to create more sharpness in the sound, compared to the Kuijken’s that preserves this for the climax case only.    


Fig. 9 – The attack slope of the Prelude audio signals from (a) Kuijken and (b) Rampal.

Regarding the sound brightness, Rampal exhibits 40.2% spectral energy above the 1200 Hz, whereas Kuijken shows 17.8% only, clearly reflecting the structural differences and tuning between the two instruments.Pitch analysis: Figures 10(a) and (b) depict the main pitch distribution of the Prelude audio signals from Kuijken and Rampal, respectively. Apparently, there is a high similarity in the organization, as both performers correctly interpret the piece; nevertheless, some differences are notable. Clearly, the difference in tuning shows a shift to lower frequencies in the case of Kuijken (e.g., the first note is A (enclosed in the green horizontal rectangular in Fig. 10) that corresponds to 405Hz instead to 440Hz as in Rampal’s case). Moreover, the ascending motion of Fig. 6 that corresponds to the pitches enclosed inside the red eclipse in Fig. 10 is more profound in the case of Kuijken rather than in Rampal’s. In addition, there are some clearer spectral lines in the case of Kuijken rather than in Rampal’s (see black square in Fig. 10).

Fig. 10 – The main pitch distribution of the Prelude audio signals from (a) Kuijken and (b) Rampal.


Figure 11 deepens in the articulation of the spectral flux by zooming at specific points of Fig. 10. In particular, Fig. 11(a) shows how the starting A note is formed by the two performers. Kuijken starts with lower pitch and gradually achieves a mean frequency of 405Hz (Fig. 11(a)-top), whereas Rampal does exactly the opposite; he starts with higher frequency around 447Hz and, via a vibrato mode, he gradually converges to a mean frequency of 440Hz (Fig. 11(a)-bottom).


Fig. 11 – Differences in the articulation of the spectral flux of Kuijken’s (top) and Rampal’s (bottom) interpretation of the (a) opening note A and (b) the sequence of the opening notes A-C-E.


Moreover, when looking at the sequence of the opening three notes (first bar in Fig. 4) expressed in the pitch domain (Fig. 11(b)), Kuijken (Fig. 11(b)-top) employs less vibrato and more legato (especially for the connection of C and E; 3-4s), whereas Rampal does exactly the opposite (Fig. 11(b)-bottom), i.e., he disconnects the three notes and gradually employs more vibrato (especially for the E; 2.9-3.8s). These differences, show how the interpretation of the performer affects the audio stimulus that arrive to the audience (see Fig. 1), following divergent ways in the interpretation of the Telemann’s compositional intention. Apparently, again Rampal’s playing seems more exocentric compared to the Kuijken’s one, as he wants to dominate his playing style from the first three notes, transferring the notion of a more contemporary approach in the Telemann’s introductory compositional gestures.   

Similarity analysis: Figures 12(a) and (b) depict the self-similarity matrices (shifted for horizontal diagonal) of the Prelude audio signals from Kuijken and Rampal, respectively. An increased similarity and periodicity is expected along the first five bars. In Fig. 12 regions of high similarity appear as bright red squares on the diagonal. These are quite similar in both interpretations, as they both reveal the Telemann’s structure of the Prelude. For example, bars 1, 3 and 5 in Fig. 4 show a high structural similarity, presenting a periodic characteristic. This is reflected in Figs. 12(a) and (b), appearing as bright red squares sequence inside the three first white boxes. As noted before, Kuijken produces clearer harmonic lines (Fig. 10(a)), implying a clearer representation of the structure of the piece. This is also seen in the fourth white box of Fig. 12, where the repetition of the sequences and the ascending motion of Fig. 6 are represented as more distinctive red squares in Kuijken’s case (Fig. 12(a): 25-30s) than in Rampal’s case (Fig. 12(b): 17-23s). The notion of repeated sections, as chain sequences are reflected in both sub-figures of Fig. 12, through the diagonal stripes or checkerboards appearing offset from the main diagonal.


Fig. 12 – The self-similarity matrices of the Prelude audio signals from (a) Kuijken and (b) Rampal.


Novelty Curve: Figures 13(a) and (b) depict the novelty curve of the Prelude audio signals from Kuijken and Rampal, respectively.


Fig. 13 – The novelty of the Prelude audio signals from (a) Kuijken and (b) Rampal.


As it is clear from Fig. 13, both Kuijken’s and Rampal’s performance exponentially maximize the novelty across the Prelude. What it is different is the point of novelty maximization; in Kuijken’s performance the novelty maximum is reached where the climax appears, whereas in Rampal’s one this is reached at the end. This shows that Rampal gradually builds the anticipation of the new material towards the end, and not towards the climax as Kuijken does, evoking further the expectation of the next part that will follow (i.e., the Fugue).

Chromagram analysis: Figures 14(a) and (b) depict the self-organizing map projection of the chromagram of the Prelude audio signals from Kuijken and Rampal, respectively. In Fig. 14(b), there is a clear concentration (red peaks) around the same tonalities identified in the music analysis and the harmonic progression depicted in Fig. 4. The differences of the peaks location between Fig. 14(a) and (b) lie into the difference in the tuning between the two instrument that affects the mapping, which is based on the well-tuned system assuming A in 440Hz, like the modern flute adopts.


Fig. 14 – The SOM projection of the chromagram of the Prelude audio signals from (a) Kuijken and (b) Rampal.


Emotional analysis: Figures 15(a) and (b) depict the three emotional dimensions (activity, valence and tension), as they evolve across Prelude audio signals from Kuijken and Rampal, respectively. Apparently, the structural characteristics of the two performances are differently reflected in the emotional space.


Fig. 15 – The three emotional dimensions (activity, valence and tension) as they evolve across of the Prelude audio signals from (a) Kuijken and (b) Rampal.

As it can be seen from Fig. 15(a), Kuijken’s performance keeps the tension (tense arousal) high almost across the whole duration of the Prelude, at a distance from valence and activity (energetic arousal) dimensions. Especially, valence is kept almost at a neutral level (considering 4.5 as the neutral value) and the activity is framed between 0 and 4. On the contrary, Rampal’s performance keeps the tension at a lower level (around 7) and reduces it towards the end, whereas it keeps the activity (energetic arousal) at a higher level (around 4) compared to the Kuijken’s performance. Finally, the valence is more positive, as mainly lies at values >4.5, exhibiting the highest value (around 7) very close to the end of the piece, where both the tension and activity are significantly reduced. The latter shows more engagement with the character of the Prelude, building an atmosphere that could establish the Telemann’s music language and transmit a positive and calm approach at the end, as a kind of memory of the previous and anticipation of the new to come.


6. Conclusive Remarks

Telemann’s Prelude, despite its short duration, was able to expose the Baroque style through the flute and managed to provide the basis for the bilateral approach explored in this paper, i.e., the music analysis and the advanced signal processing one. The identified structural characteristics at the score level impose different qualities at the auditory level that reveal Telemann’s intentions, as they are materialized via the two different interpretations by eminent performers of flute traverso and modern flute, respectively. The goal was to realize how the compositional intentions are transformed in a language that is mostly understood by the musicians and, at the same time, are codified in the audio signal that reaches the listener in a way, which reveals their full spectrum. To achieve this, many levels of analysis were employed, including duration, dynamics, pitch, timbre, self-similarity, novelty, chroma and predicted emotional impact. This concept allowed for the examination of the micro/macro-structural events of the Prelude in a time-dependent representation, following, thus, the dynamic flow of the music in parallel to the changes in the modeling parameters explored here. The bilateral perspective introduced in this paper addresses the need for more efficient approaches of the compositional thinking and intensions through the analysis of both the music score and the audio output. As the latter is the one that directly reaches the audience, perhaps this spherical perspective could bridge the compositional intentions reflected in the abstract forms of homogeneity, discontinuity, correlation, with the actual musical content.    



[1] MFCC is a representation of the short-term power spectrum of a sound, based on a linear cosine transform of a log power spectrum on a nonlinear MEL scale of frequency, which approximates the human auditory system's response more closely than the linearly-spaced frequency bands used in the normal cepstrum. This frequency warping can allow for better representation of sound, for example, in audio compression.
[2] A self-organizing map (SOM) or self-organizing feature map (SOFM) is a type of artificial neural network (ANN) that is trained using unsupervised learning to produce a low-dimensional (typically two-dimensional), discretized representation of the input space of the training samples, called a map, and is therefore a method to do dimensionality reduction. SOMs differ from other artificial neural networks, as they apply competitive learning as opposed to error-correction learning (such as backpropagation with gradient descent), and in the sense that they use a neighborhood function to preserve the topological properties of the input space.



This paper was presented at 6th EIMAD – Meeting of Research in Music, Art and Design, and published exclusively at Convergences.


Bibliographic References

Brown, R. (n.d) “Telemann Fantasias: a feat of ingenuity and inspiration”, available at:

Eerola, T., Lartillot, O., and Toiviainen, P. (2009). Prediction of multidimensional emotional ratings in music from audio using multivariate regression models. International Conference on Music Information Retrieval, Kobe, 2009.

Foote, J., Cooper, M., and Nam, U. (2002). Audio retrieval by rhythmic similarity. Proc. of the 3rd International Conference on Music Information Retrieval, Paris. Available at

Juslin, P. N. (2000). Cue utilization in communication of emotion in music performance: relating performance to perception. Journal of Experimental Psychology: Human Perception and Performance, 26(6), 1797-813.

Kuyjken, B. (1987). Georg Philipp Telemann, 12 Fantasias for Flute, Wiesbaden: Breitkopf & Härtel.

Porter, A., Bull, C., and Pyle, D. (2008). Telemann: 12 Fantasias for Flute without bass. A study guide with amy porter. DVD produced by Mike Wilkinson, (Michigan: Duderstadt Media Center). 

Schulenberg, D. (2008). Music of the Baroque, New York: Oxford University Press, 249. 

Toiviainen, P. and Krumhansl, C. L. (2003). Measuring and modeling real-time responses to music: The dynamics of tonality induction. Perception, 32(6), 741–766.

Zohn, S. (2008) Music for a Mixed Taste: Style, Genre and Meaning in Telemann’s Instrumental Works, New York: Oxford University Press, 337. 

Reference According to APA Style, 5th edition:
& Dias, S. Hadjileontiadis, L. (2018) From Telemann’s compositional ideas to performance quali-ties and audio footprint signal processing analysis. The paradigm of Telemann’s Prelude from Fantasia Nº 2 à Travers sans Basse. Convergências - Revista de Investigação e Ensino das Artes , VOL XI (21)