Autonomous detection of neotropical sciaenid fishes

Sciaenid passive acoustics are a demonstrated valuable tool for fisheries management. In spite of this, an efficient software tool to detect and identify fish sounds is not currently available. Such tool would be useful for autonomous recognition and array methodologies. For Neotropical environments this lack is even more conspicuous since the availability of corroborated sciaenid sounds is limited. We are developing such tools using corroborated Cynoscion squamipinnis (Pisces: Sciaenidae) sounds. Our approach is based on timbre statistics, short and long-term partial loudness, and the 30 Hz typical pattern found on the signal's stridulations. Relevant fish drums are detected through empirically found fix thresholds for the timbre statistics and the 30 Hz pattern, and a dynamic threshold established by an unsupervised algorithm based on the long-term loudness. Current results show a recognition rate of 80%. Despite these promising numbers, there are still challenges ahead. In the future, we plan to incorporate other variables that affect underwater sound characteristics such as depth, source level distance, and physical chemical properties, which may be crucial to make a user friendly, accurate, and practical tool, for neotropical marine environmental managers. We also plan to extend this method to other soniferous coastal fish.