¡@¡@Taiwan is surrounded by sea, there are coral reefs at both south and north ends, and also at off-shore islands. Coral reef offers habitat and resources for marine animals, that are attracted to in numbers, however due to the limited resources, competitions and conflicts are common among animals. In order to defend the territory on intimidate intruder, some fishes develop vocal mechanism that certain sounds are generated at encounter. The sounds of common Holocendridae fishes at coral reef were studied before, but the sound characteristics of sound were not fully investigated. In this research, sound samples were recorded from Neoniphon sammara, Myripristis murdjan, and Sargocentron spinosissimum(Holocentridae) in an indoor water tank, to understand more about their specific features of sounds generated at disturbance. The identification system of this study was based on Matlab, which extracted the characteristic parameters from the sounds, so the database for comparison can be formed. The sounds can be classified into single pulse and pulse train, and the parameters used are dominant frequency, band width, duration, and pulse repetition rate. As the result, Neoniphon sammara has single pulse frequency of 428.0¡Ó95.0 Hz (mean ¡Ó standard deviation), pulse train frequency of 449.8¡Ó92.4 Hz, Myripristis murdjan has single pulse and pulse train frequency of 375.2¡Ó96.3 Hz and 369.2¡Ó96.0 Hz, and Sargocentron spinosissimum has single pulse and pulse train frequency of 377.6¡Ó93.5 Hz and 387.8¡Ó97.1 Hz. The similarity of sounds between Neoniphon sammara and Myripristis murdjan is 81.3%, between Myripristis murdjan and Sargocentron spinosissimum is 83.6%, and between Neoniphon sammara and Sargocentron spinosissimum is 90.3%. Finally, the identification accuracy of Neoniphon sammara is 78.9%, Myripristis murdjan is 71.4%, and Sargocentron spinosissimum is 38.4%. The low identification accuracy of Sargocentron spinosissimum is due to the high similarity of sounds with Myripristis murdjan, and the number of sound samples is not sufficient.

In the study of sound identification, land animals such as birds and bats have been well investigated, and so are their habitats. On the other hand, sound making creatures in the ocean are much less researched. In this research, the stress calls of three Holocentridaes, Neoniphon sammara, Myripristis murdjan, and Sargocentron spinosissimum, who are commonly found in coral reefs, were recorded in water tank for analysis of sound characteristics. The averaged characteristic parameters of single pulse among three is around 410 Hz for the peak frequency, 100 Hz for the bandwidth, 0.07 dB/Hz for the slope, and duration of 0.05 s. As for the impulse train, averaged peak frequency is 415 Hz, 55 Hz for the bandwidth, 0.07 dB/Hz for the slope, and duration of 0.5 s. These parameters were first checked by the Kolmogorov-Smirnov Test to identify if each parameter follows normal distribution; the slopes of ascending and descending frequency and the total duration are not in normal distribution. The three parameters were later transferred so as to concentrate variances. Next, analysis of variance was applied on all characteristics to extract the significant parameters (including non transferred and transferred data), which were then tested by Stepwise Discriminat and Back-propagation Network. The identification rate of for single pulse with and without data transfer is 63% and 82% while pulse train is 57% and 73%. Both identification rates were raised up approximately 20% due to the data transfer. Both methods provide an reliable tool for marine sound identification, and the whole process of the study may be applied to another biological identification.