ELEC-C5220 - Koneoppiminen informaatioteknologiassa, Luento-opetus, 11.1.2024-25.4.2024
This course space end date is set to 25.04.2024 Search Courses: ELEC-C5220
Project instructions
Overview and timeline
Total points: 20 points from final report and 20 combined points from Milestone 1 and evaluation
Milestone 1 (15pt)
Building a baseline denoising system and passing related unit tests awards up to 15 points.
Deadline 22.4.
Evaluation metrics (5-20pt)
Submit de-noised test set files for evaluation. Getting close enough to baseline performance awards full marks and replaces any missing points from Milestone 1.
Evaluation system will be open from 12.4. to 29.4.
Project report (20pt)
Up to 5 pages in pdf format, returned to MyCourses.
Deadline is 29.4.
Project Milestone 1
Deadline 22.4.2024 at 23:59
This project milestone contains a set of unit tests for a reference implementation of a speech denoising system. Submitting your code with this notebook awards up to 15 exercise points.
The reference implementation is intended as a support and guide. Feel free to experiment or even use tools other than PyTorch. You can find the project reference implementation template at https://version.aalto.fi/gitlab/elec-c5520-students/elec-c5220-project-2024/
Denoising system evaluation
Evaluation system will be open from 12.4. to 29.4.
Full marks (40 points) for the project are available for any system that is
- clearly detailed in the final report and
- can reach good enough test set Log-mel-spectrum distance in the evaluation notebook
In other words, if your system reachers high enough test set performance, you will get 20 points from the evaluation and can ignore any missing points from Milestone 1.
Each group can download the test set samples, apply their method, and return the denoised samples for evaluation. and there will be no limit on the number of evaluations. Last returned submission counts.
Project report (20pt)
Project report deadline is 29.4.
Return a written report describing your project system and results. There will be a MyCourses return box for this.
You can use the following LaTeX templates as a starting point. Page limit is 5 pages including references, but there is no obligation to fill all five.
Finnish report template https://www.overleaf.com/read/cyhmdywspgrw#f9b53e
English report template https://www.overleaf.com/read/mhffqqvfywhq#ce50c4
Accessing project data on other computers
When working on JypyterHub you can find some speech and noise samples in the`$COURSEDATA` directory
- Speech samples are in `$COURSEDATA/project-data/ARCTIC_elec_c5220_project`
- Noise samples are in `$COURSEDATA/project-data/noise_dataset`
You can also download these samples for your local machine or access them on other Aalto computers. Use the course slug `koit2024` and follow the instructions below