We conducted a laboratory study in the facilities provided by the Natural History Museum of Funchal (Mosquito Lab). Three species of mosquitoes were recorded to determine their dominant frequencies and spectral behaviors. The species used for this collection and study were A. Aegypti, C. Quinquefasciatus & Pipiens and Culiseta, which came from a lab colony established from captures collected in Funchal city in 2019.
The mosquitoes were kept in an environmental room simulating natural conditions, with 60 ± 10% relative humidity and temperature of 20–25 °C. Mosquitoes were housed individuals in boxes (25 × 25 × 25 cm) covered with a mesh cap. They were fed with 20% sucrose solution supplemented with 1 g aquarium fish food mixed daily from the brand “Sera Guppy Gran”. The duration of the study was approximately 48 days. All mosquitoes used in these experiments were 7–25 days old. For the recording process, sensors were incorporated into the boxes and the tests conducted on 12–18 specimens for Aedes Aegypti, 7–12 specimens for Culex and 4 specimens for Culiseta. The duration of the extracted sequences ranged from 0 to 300 ms. To generate samples closer to real-world acquisition conditions we added environmental noise in some mosquito samples.
Uncompressed audio of real sound waves was converted to digital format without any further processing. This means that recordings are exact copies of the source audio, recorded in WAV files.
The acoustic sensor uses a low-noise omnidirectional microphone capsule2. The microphone converts sound into electrical signals with a specific signal to noise ratio (80 dB), self-noise, and residual noise. All these parameters influence the quality of the acquired sound.
Noise can be a significant problem when acquiring physical signals as voltages. Signal smoothing attempts to capture the essential information in the signal while leaving out the noise. This is done by interpolating the raw signal to estimate the original one17.
To collect samples, we used three devices: one of them was our prototype comprising a Teensy 3.2 audio board, microphone and environmental sensor for 44.1 kHz sampling rate. The other two were general-purpose smartphones (Huawei P20 Lite and IPhone 4) used to record samples with a 8 and 48 kHz sampling rate, respectively.
To start a colony for our experience, we installed traps and buckets of water to catch eggs and adult mosquitoes. The female Aedes mosquitoes require a blood meal before each egg-laying18. The eggs are deposited individually on the inner walls of any container capable of storing water. This work was conducted jointly with the Natural History Museum of Madeira and IASaude (the regional health authority of Madeira islands) as part of a plan to control the spread of mosquitoes in the city of Funchal (Fig. 1).
Location and number of traps in the city of Funchal, Madeira, Portugal.
A. Aegypti mosquitoes, lay the most eggs in the velcro tape, while Culex and Culiseta prefer to lay directly in rafts on still water or in other substances19. Traps with a ventilation system were also used to capture adult mosquitoes, especially Culex and Culiseta.
Figure 2 shows the procedure from egg collection to mosquito germination, and also the boxes that are used for further acquisition of sound samples. It is noteworthy that after 25–30 days the mosquitoes die due to the conditions imposed in the study.
Procedure for collecting audio samples for different species.
Step A comprises the gathering of eggs and mosquitoes. The figures show a bucket inside which mosquitoes lay eggs on a velcro tape, and also a trap. These traditional methods allow a fine assessment of the distribution of mosquito populations over time and space (periodically summarized in epidemiological bulletins). In step B the collected eggs are germinated to create a colony. Then, (step C) mosquitoes are placed in boxes and fed with a sugar solution and fish food20. Finally, in step D, audio samples are collected by the devices: mobile phones and low-cost IoT. This procedure is repeated when the colony dies after 25 days, starting from step B.
Audio was recorded inside boxes (25 × 25 × 25 cm) where the mosquitoes were located at a maximum distance of 27 cm from the microphone placed in the center of the box. The signal amplitude fluctuates significantly over time as the mosquitoes in free flight approach the microphone or move away.
Continuous recordings were then split into 300 millisecond (ms) snippets. Since mosquitoes have a very short flight, it was necessary to apply a slight stimulus on the wall of the boxes (covered by a net) to force them to fly.
To analyze each mosquito recording, 34 features were extracted taking into account several parameters of the signal belonging to three different domains: time (1–3), frequency (4–8, 22–34) and cepstrum (9–21), analyzed below in the Technical Validation section21,22.
These features are often used for speech signal classification, but are useful when handling non-speech signals as well. They enable a comprehensive analysis of the mosquito sounds in terms of amplitude, energy, zero crossing rate, power, frequency variation in the audio file, tonality, loudness, etc. The features are included in the dataset23 and their computation is demonstrated in the Code Availability section.
Source: Ecology - nature.com
