monkeycall 0.0.4

Detect (spider/howler) monkey calls using pretrained TF2 models from Hugging Face.

Monkey Call Detector

Detect spider monkey whinny calls (and howler monkey calls) in audio recordings using pre-trained TensorFlow models.

This package downloads a model from 🤗 Hugging Face and provides simple functions for ecologists and data scientists to run inference on short clips or entire recordings.

---

Installation

This has been tested with a fresh python 3.13 installation.

(Optional) Make a virtual environment (python -m venv /path/to/new/virtual/environment)

pip3 install monkeycall  # This will also install requirements.

Usage examples

You can now detect monkey calls. To detect spider monkeys try the following:

Detecting the Spider Monkey whinny

from monkeycall.spider_monkey import load_model, predict_recording

recording_path = "/path/to/sample.WAV"

repo_id = "georgiosrizos/spider-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Reading long recording:", recording_path)
output_df = predict_recording(audio_path=recording_path,
                              model=model,
                              hop_size=1)

print("The recording has been clipped in 3 sec window clips. The whinny detection probabilities are:", output_df)

# You can save the pandas DataFrame to review later.
output_df.to_csv("predictions.csv", index=False)

Or if you already have a 3 sec clip:

from monkeycall.spider_monkey import load_model, predict_3_sec_clip

recording_path = "/path/to/3sec.WAV"

repo_id = "georgiosrizos/spider-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Reading clip:", negative)
output = predict_3_sec_clip(audio_path=recording_path,
                            model=model)

print("The probability this sample contains a whinny is:", output)

If you have a folder containing multiple audio files, you can run the following:

from monkeycall.spider_monkey import load_model, predict_recording

input_folder = "/path/to/audio/file/folder"
output_folder = "/path/to/store/positive/clips/folder"

repo_id = "georgiosrizos/spider-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Search folder for calls:", long_recording_folder)
output_df = search_folder_for_calls(input_folder=long_recording_folder,
                                    model=inference_fn,
                                    hop_size=1,
                                    probability_threshold=0.5,
                                    output_folder=output_folder,
                                    extensions=(".wav", ".mp3", ".flac", ".ogg"))

print("The recordings have been clipped in 3 sec window clips. The whinny detection probabilities are:", output_df)

# You can save the pandas DataFrame to review later.
output_df.to_csv("predictions.csv", index=False)

Detecting the Howler Monkey hoot and roar

from monkeycall.howler_monkey import load_model, predict_recording

recording_path = "/path/to/sample.WAV"

repo_id = "georgiosrizos/howler-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Reading long recording:", recording_path)
output_df = predict_recording(audio_path=recording_path,
                              model=model,
                              hop_size=2)

print("The recording has been clipped in 5 sec window clips. The hoot and roar detection probabilities are:", output_df)

# You can save the pandas DataFrame to review later.
output_df.to_csv("predictions.csv", index=False)

Or if you already have a 5 sec clip:

from monkeycall.howler_monkey import load_model, predict_5_sec_clip

recording_path = "/path/to/5sec.WAV"

repo_id = "georgiosrizos/howler-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Reading clip:", negative)
output = predict_5_sec_clip(audio_path=recording_path,
                            model=model)

print("The probability this sample contains a hoot is:", output[0])
print("The probability this sample contains a roar is:", output[1])

If you have a folder containing multiple audio files, you can run the following:

from monkeycall.howler_monkey import load_model, predict_recording

input_folder = "/path/to/audio/file/folder"
output_folder = "/path/to/store/positive/clips/folder"

repo_id = "georgiosrizos/howler-monkey-detector-SEResNet"

print("Reading pre-trained model.")
model = load_model(architecture=repo_id)

print("Search folder for calls:", long_recording_folder)
output_df = search_folder_for_calls(input_folder=long_recording_folder,
                                    model=inference_fn,
                                    hop_size=2,
                                    probability_threshold=0.5,
                                    output_folder=output_folder,
                                    extensions=(".wav", ".mp3", ".flac", ".ogg"))

print("The recordings have been clipped in 5 sec window clips. The hoot and roar detection probabilities are:", output_df)

# You can save the pandas DataFrame to review later.
output_df.to_csv("predictions.csv", index=False)