Jon Nordby - Sound Event Detection with Machine Learning
Sound Event Detection with Machine Learning
[EuroPython 2021 - Talk - 2021-07-29 - Parrot [Data Science]]
[Online]
By Jon Nordby
Sound Events (or Audio Events or Acoustic Events) are individual distinct sounds. This could be the pop of roasting popcorn kernels, the cough of a patient, a car that is passing by on a road, or the sound of an alarm in an office building. Sound Event Detection (SED) is the task of detecting such sounds, returning precise times that each kind (class) of sound occurs. It finds uses in music analysis, manufacturing, medicine and noise monitoring.
We will show how to realize a basic Sound Event Detection system in Python, using fermentation tracking of beer brewing as a fun and practical example. The talk will cover all major parts of such a system, including:
Collecting and exploring a custom dataset
Setting up the supervised learning task from the dataset
Extracting spectrogram features from audio waveforms
Training a Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN)
Running the trained model on an real-time audio stream
Processing model output probabilties into discrete events
Evaluate the performance of the resulting SED system
Example code in Python covering these aspects will be provided.
Libraries used with be Keras, TensorFlow and scikit-learn for machine learning, and pysoundfile, sounddevice and librosa for audio processing, with some numpy and pandas for general data manipulation.
License: This video is licensed under the CC BY-NC-SA 4.0 license: https://creativecommons.org/licenses/...
Please see our speaker release agreement for details: https://ep2021.europython.eu/events/s...
Смотрите видео Jon Nordby - Sound Event Detection with Machine Learning онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь EuroPython Conference 11 Октябрь 2021, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 3,862 раз и оно понравилось 52 людям.