Everyone thought the stupid Brit had faded away, well…not so fast
I was disappointed to discover that my tame beekeeper was more interested in Gott über den Bienen so my acoustic gathering project got undermined by propolis - gods work, truly, it was quite distinctive in the recordings unsurprisingly
now that AI is taking leaps and bounds courtesy of Microsoft and Google should we reconsider how this could be used in an open source community manner?
I asked ChatGpt4 its opinion on my previous efforts - well! now I discover I was climbing the impossible beehive!
Q: Can you suggest a neural network model for generating categorizing responses from a mems microphone recordings of bees?
Yes, I can suggest a neural network model for generating categorizing responses from a MEMS microphone recordings of bees. One possible approach is to use a convolutional neural network (CNN) for audio classification.
Here is a possible architecture for the CNN:
Input layer: The input layer should be designed to receive the audio signals recorded by the MEMS microphone.
Convolutional layers: The convolutional layers will extract features from the audio signals by applying a set of filters to the input data. These filters are learned during training and allow the network to recognize different patterns in the audio.
Pooling layers: The pooling layers will reduce the spatial dimensions of the data and help the network to be more robust to variations in the input.
Flatten layer: The flatten layer will reshape the output of the convolutional layers into a 1D vector, which can be used as input to the fully connected layers.
Fully connected layers: The fully connected layers will perform the final classification of the audio signals. These layers take the features extracted by the convolutional layers and use them to predict the corresponding category of the audio.
Output layer: The output layer will produce a probability distribution over the possible categories of the audio signals.
To train the CNN, you will need a labeled dataset of audio recordings of bees. You can use this dataset to train the network using a supervised learning approach, where the network learns to predict the correct category of the audio signals based on the input features and the corresponding labels.
Once the CNN is trained, you can use it to classify new audio signals recorded by the MEMS microphone. The network will output a probability distribution over the possible categories, allowing you to identify the most likely category of the audio signal.
Q: What should the labels be to get a robust set of responses?
To get a robust set of responses, the labels for your dataset should be accurate and representative of the different categories of bee sounds that you want to classify. Here are some suggestions for labeling your dataset:
Species of bees: If you are interested in distinguishing between different species of bees, you could label your dataset with the corresponding species names.
Behaviors of bees: Bees make different sounds depending on their behaviors, such as foraging, mating, and communication. Labeling your dataset with these behaviors can help the network to identify the corresponding sound patterns.
Environmental factors: Environmental factors, such as temperature, humidity, and light, can affect the sounds that bees make. Labeling your dataset with these factors can help the network to learn how different environmental conditions affect the sounds of bees.
Health of bees: The sounds that bees make can also provide information about their health, such as whether they are infected with a disease or parasite. Labeling your dataset with these health conditions can help the network to identify patterns in the sounds that are indicative of different health states.
It’s important to ensure that the labels in your dataset are consistent and accurately reflect the different categories of bee sounds that you want to classify. Additionally, you should try to include a diverse range of examples for each category in order to train a robust model that can generalize well to new data.
Q: How can I simplify the labeling or data set to predict swarming?
Swarming is a complex behavior of bees that involves a number of different factors, including colony size, age of the queen bee, availability of food and water, and environmental conditions such as temperature and humidity. Therefore, it may be difficult to simplify the labeling or dataset for predicting swarming behavior. However, here are some suggestions that could potentially help:
Focus on specific audio features: Swarming behavior is often accompanied by a characteristic buzzing sound that is produced by the bees. By focusing on specific audio features that are indicative of this buzzing sound, you may be able to simplify the labeling of your dataset. For example, you could label your dataset based on the presence or absence of a specific frequency range or modulation pattern that is associated with swarming behavior.
Use clustering techniques: If you have a large dataset of audio recordings, you could use clustering techniques to group together similar recordings that are indicative of swarming behavior. This could help to simplify the labeling process by reducing the number of distinct categories that need to be labeled manually.
Gather data during swarming season: Swarming behavior typically occurs during a specific season or time of year, depending on the location and climate. By gathering data specifically during swarming season, you may be able to reduce the variability in your dataset and simplify the labeling process.
Collaborate with bee experts: Finally, it’s important to remember that swarming behavior is a complex phenomenon that is influenced by a wide range of factors. Therefore, it may be helpful to collaborate with bee experts who can provide insights into the behavior and help to guide the labeling process. Beekeepers, entomologists, and other experts in the field may be able to provide valuable input into the design of your dataset and the labeling process.
Q: Can you suggest the frequencies that would indicate swarming?
Swarming behavior in bees is often accompanied by a characteristic buzzing sound, which is produced by the bees as they fly around in a cluster. The frequency range of this sound can vary depending on a number of factors, such as the size of the swarm and the distance from the swarm. However, here are some general frequency ranges that have been observed in swarming bees:
Low-frequency sounds: Swarming bees often produce low-frequency sounds in the range of 100-400 Hz. These sounds are typically produced by the beating of the bees’ wings, and can be heard from a distance of several meters.
High-frequency sounds: In addition to low-frequency sounds, swarming bees may also produce high-frequency sounds in the range of 2-10 kHz. These sounds are thought to be produced by the vibration of the bees’ wings against their bodies, and may be more difficult to hear from a distance.
Modulated sounds: Swarming bees may also produce sounds that are modulated or vary in frequency over time. For example, the frequency of the buzzing sound may increase or decrease as the bees move around in the swarm.
It’s important to note that the frequency ranges associated with swarming behavior can vary depending on the specific species of bees and the environmental conditions. Additionally, other factors such as background noise and the proximity of the microphone to the swarm can also affect the frequency range of the recorded sounds. Therefore, it may be necessary to experiment with different frequency ranges and signal processing techniques to accurately detect swarming behavior from audio recordings.
Am thinking of taking up this fascinating work, put my money where my mouth is…plus bees are the worlds’ workers doing God’s work [as my neighbour should know].
Any future in AI and beehives?
1 Like
I would not count on all things written in the internet and so particular not on bee biology expert knowledge chatty (not?) has. There are a lot of frequency assumption – what happens in the bee hive at which dominat frequency – especially a time ago from people selling a bee sound app. I would doublecheck the facts so that we do not spend time on the wrong place.
So I would definitive count on this recommendation:
Collaborate with bee experts
:-)
1 Like
Hey John, I’m also trying to figure out how to measure/analyze sound data (ML). I’d love to connect and maybe brainstorm together. Having read the peer lit. on honeybee sound, it’s clear to me that we’re not sure where to position microphones within a hive and if there are differences based on this … so that’s where I’m starting. Can we chat?
mois
July 10, 2023, 6:22am
5
maybe even better: quadruple-check! because LLMs are producing mistakes by design (because they pick random tokens every once in a while as described by stephen wolfram in his essay “What Is ChatGPT Doing … and Why Does It Work?” ).
and if it has to be LLM-ai, we should start to check out LAION, a public non-profit organization (e.V.) from hamburg/germany, that
“provides datasets, tools and models to liberate machine learning research. By doing so, we [they; mois] encourage open public education and a more environment-friendly use of resources by reusing existing datasets and models.”
