Acoustic investigation of chocolate by changing temperature and cocoa content

In the experiments, the sound generated during the breaking of chocolate samples was examined. The fracture was performed by a precision penetrometer, the breaking sound was recorded. Texture index (TI) was calculated from the resulting signal. First the change of the resulting TI was monitored as a function of the samples ’ temperature. The sample groups of the same dark chocolate with different temperatures were completely separated from each other with statistical tool (LDA, linear discriminant analyses), but no trend was found to describe the change. Secondly, based on the TI, we could identify and classify the chocolate samples in the appropriate groups (based on cocoa content from 40 to 85%). According to linear discriminant analyses chocolates with different cocoa content were completely separated and showed a certain pattern. Based on the obtained results, it can be stated that the cocoa content of chocolate can be determined on the basis of TI obtained by acoustic method.


INTRODUCTION
The food consumed in our meals is experienced with all our senses. A lot of food gives a certain sound -like crunch, snap, crack, etc.when consuming it. The idea that sound sensation plays an important role in the perception of textures was first published by Drake (1963), thereby promoting food texture analysis. If a crunchy product does not induce the expected sound after the bite, it can be attributed to the texture, the poor quality or inadequate components or process. W ojtowicz et al. (2015) investigated the effect of various recipes and processing conditions of breakfast cereals on the finished product. From the results obtained, the appropriate manufacturing technology was determined. In their study of potato chips, Salvador et al. (2009) concluded that the force/deformation test and the acoustic information received at the same time could be an effective tool for comparing sensory results in crisp products. Mohamed et al. (1982) also demonstrated in their study of different waffles for different influences that there is a significant correlation between sensorial sound characteristics and instrumental measurements (r 5 0.701). Vickers (1976) found that combinations of acoustic and mechanical tests describe food sounds better than any technique alone in the case of different crispy food materials. Two main approaches are used to test the fracture's sound, either the amplitude-time curve of the acoustic signal (Jakubczyk et al., 2017, Gondek et al., 2013, Salvador et al., 2009 or the amplitude frequency curve is taken up and is used for further analyses. Using a microphone Jakubczyk et al. (2017) tested various coated and/or stuffed extruded products. The force/ deformation test was performed with a TA-HD plus texture analyser. In the experiments of Taniwaki and Sakurai (2008), the food structure is quantitatively evaluated based on the texture index (TI), which is defined as energy density. Using texture measurements is widely spread to describe quality of chocolate, as for example was done by Biczo et al. (2013).
In our work, we aimed at compiling a mechanical-acoustic method suitable for examining chocolate products. A further objective was to investigate the effect of the sample temperature during the test and the effect of the chocolate cocoa content on the resulting sound response in dark chocolate and milk chocolate samples.

MATERIALS AND METHODS
For the temperature-dependence tests, 50% cocoa containing dark chocolate by Fin Carr e (Lidl private label chocolate with a fat content of 52%) was used. Chocolate samples were placed in a fridge (þ48C) for 1 hour. Then, the temperature of 8 samples was measured every 3 minutes with an infrared thermometer and the acoustic response of their breaking was recorded, and the remaining samples were measured every 5 min. Three parallel measurements were made on the same batch of chocolates (a total of 39 samples). From the measured temperatures, 5 temperature groups were formed (Group 1 -8-10 8C, Group 2 -13.1-16 8C, Group 3 -16.1-19 8C, Group 4 -19.1-22 8C and Group 5 -22.1-25 8C), in which samples were classified.
For the cocoa content test, samples were provided by Interfood & Pharma Kft. The chocolates were of the same shape, with varying cocoa contents, 20 samples for each cocoa content were tested. Chocolate cubes (25 mm*20mm*7mm) were placed in a fridge for one day before the acoustic tests and their temperatures were recorded directly before measuring. Six types of dark chocolate and 2 types of milk chocolate were examined (a total of 160 samples). The list of different types and cocoa contents of the chocolates is shown in Table 1.
The sound of the chocolate breaking was recorded using a microphone in the set up shown in Fig. 1 and recorded with a 'WavePad' program (WavePad Sound Editor v 5.02, NCH Software) at 44.1 kHz sampling frequency. The chocolate pieces were supported by a ring-shaped sample holder. For the breaking SMS TA-XT2 Precision Penetrometer (by Stable Microsystems Ltd.) with spherical probe was used, penetration depth was 7 mm and the rate of speed was 1 mm/s. Then the data from the scanned signal were normalised to always be between -1 and 1, and then they were analysed using the Fast Fourier Transform (FFT).
The frequency-domain data (20 Hz-20 kHz range) were then sub-divided into third-octave intervals and the TI was calculated for each frequency band:  Progress in Agricultural Engineering Sciences 16 (2020) S1, 115-122 where f L represents the lowest and f H the highest frequency of each frequency band (third-octave intervals), -V i is the amplitude of the texture signal, and n is the number of data points in the given frequency band.
The processing of the signal and the calculation of the 'Texture Index' were done in the Mathcad program based on the algorithm developed by the Department of Physics and Control, SZIU. In further analyses this parameter was used to characterise the acoustic response of the chocolate's fractioning.
The statistical analyses of the data were performed by using Rstudio (R-project 3.4.1) statistical program. partial least squares regression (PLSR) analysis was used to find the correlation between the cocoa content and the frequency. Linear Discrimination Analysis (LDA) was applied to separate groups in the various measurements for acoustic sound characteristics.

RESULTS AND DISCUSSION
In the first experiment, the effect of the temperature on the acoustic response of chocolate fractioning was investigated. On the basis of our results, samples in the higher temperature groups got higher 'texture index' values. Linear discrimination analysis (LDA) showed clearly (Fig. 2) that chocolate samples of different temperature groups could be distinguished by their sound signal. However, we did not find a model which could describe consequently the change in the acoustic signal and the temperature.
In the second experiment, different cocoa-containing samples were tested for acoustic sound response. Using the results obtained by the acoustic measurements and the calculated texture index we were able to identify and classify the chocolate samples corresponding to their cocoa content with 100% certainty. The Linear Discrimination Analysis (LDA) diagram (Fig. 3) illustrates well that different chocolate cocoa-containing chocolates are nicely separated from each other and even show some pattern.
The separation of the first group (milk chocolates) is obvious because their composition differs significantly from dark chocolates, as they contain milk, which softens their texture and are therefore more flexible. Both hardness and flexibility are of great significance to the sound intensity of the chocolates, as if energy is released quickly, high intensity sounds will be generated. Thus, the hard texture with low degree of flexibility (less elastic) is more likely to provide rapid release after breaking and thus high intensity sound output. The more elastic nature allows greater deformation of the samples throughout the break, resulting in slower energy release (Gregerson et al., 2015).  We do not know exactly the reason for the difference between Group 2 and Group 3. There is no information about the exact composition of the samples; but the fat composition of cocoa beans grown at different locations may differ, which affects the properties of the chocolates. PLSR regression was performed, the results of which are shown in Fig. 4. Separately, PLSR was performed only on the dark chocolate groups (Fig. 4a) and on all samples used for the cocoa content test (Fig. 4b). In both cases, a significant relationship was found between the cocoa content and the sound response of the breaking.
The frequency at which the largest value of the regression vector was given was 68 Hz for only dark chocolates and 217 Hz for all chocolate samples together. All chocolate samples analysed together gave a higher R 2 value (0.8397) and validation was also better (R 2 5 0.7218) compared to the analysis of dark chocolate samples only. There was in any case close relationship between the cocoa content and the crispness of the chocolate samples, which should be taken care of, as the sound released during breakage plays an important role in consumers' perception of different food products, especially for chocolate.

CONCLUSIONS
Breaking sound of chocolate is an important parameter for judging the eating experience. A well-tempered, high cocoa containing chocolate is expected to break with a snap.
From our experiments we can conclude that measuring the acoustic signal of chocolates' breaking is suitable for testing the chocolates' texture. With the help of calculated texture index, sample groups of different temperatures could be distinguished, but we did not find a trend that describes the change.
Dark chocolate samples (50% cocoa content) of different temperatures gave different texture index values (higher temperature samples had higher texture indexes), and although the temperature-groups could be distinguished 100% correctly by linear discriminant analysis, no correlation could be described between temperature and texture index. Using texture index (based on the acoustic response values) we could identify the chocolate samples in the corresponding cocoa-containing group 100% correctly (from milk chocolate samples to dark chocolate samples, i.e. 54-65% cocoa and 70-85% cocoa).
Linear discriminant analyses distinguished among different chocolate samples based on their cocoa content and a certain pattern was recognisable. There were three directions according to the cocoa content, which divided our data into three groups. Group 1: milk chocolates (40%, 45% cocoa content), Group 2: chocolates with 54-65% cocoa content and Group 3: dark chocolates with 70-85% cocoa content.
These results are certainly inspiring and further studies are required as they provide a possible new measurement method. Using acoustic response, a quick method for estimating the cocoa content of chocolate could be established, with suitable equipment and the support of a statistical model. This new method would not require any sample preparation, and it was based on purely physical parameters instead of chemical characteristics.