Single Channel Source Separation with Deep Model: Tackle the Chime3 challenge
- Status
- Open
- Type
- Master Thesis
- Announcement date
- 11 Mar 2015
- Mentors
- Research Areas
Signal Processing and Speech Communication Laboratory
Diploma/Master’s Thesis: Single Channel Source Separation with Deep Model: Tackle the Chime3 challenge
Short Description :
Assume a single-channel multiple (two) speaker recording. Speech separation for such tasks can be formulated as binary classification problem in the time-frequency domain. The aim within this project is to apply develop mutli-label classification or deep learning algorithms for solving the problem. You will use deep neural networks to solve this ill-posed optimization problem. By implementing fast and reliable neural network models on a GPU and will get a broad knowledge of machine learning. If possible the outcome of your work
will contribute to a machine learning paper. We will use the model to evaluate data of the CHIME3 challenge. Therefore you will join an international competition (http://spandh.dcs.shef.ac.uk/chime_challenge/). If your are interested in this fascinating field of science simply drop me an email.
Your Tasks :
- extend a neural network model in python on the GPU using THEANO [1]
- analyze the implemented systems in terms of accuracy and performance
- contribute to scientific work in form of a paper
Your Outcome :
- learn to implement and simulate very fast Neural Networks on a GPU
- learn how to solve difficult object recognition tasks
- get a broad education in on applied machine learning
- international competition (therefore job offers afterwards possible)
Your Profile :
- motivation and reliability are a prerequisite
- good knowledge in machine learning and neural networks (at least >2 machine learning courses)
- knowledge in python programming
Additional Information :
This thesis project is planned for a duration of 6 months starting immediately. As it is a valid
contribution to an ongoing research project at the SPSC, it is rewarded with 2640e (440e per
month) and a good chance for publications.
Contact :
Matthias Zoehrer (matthias.zoehrer@tugraz.at or +43 (316) 873 - 4385)
References
Farley, and Y. Bengio, "Theano: a CPU and GPU math expression compiler," in Proceedings of
the Python for Scientific Computing Conference (SciPy), Jun. 2010, oral Presentation.
Signal Processing and Speech Communication Laboratory (SPSC), Graz University of Technology, Inffeldgasse 16c, 8010 Graz, Austria, http://www.spsc.tugraz.at
created March 10, 2014