Implement algorithms for sound processing (MFCC / Mel-frequency Cepstrum Coefficients) in an open-source framework

I am hiring developers to add support for MFCC features in the [login to view URL] Framework ([login to view URL]).

[login to view URL] is a framework for machine learning, signal processing, statistics, and other tasks related to scientific computing in .NET.

That's a complete project, comprised of at least:

1) Writing (or leverage from another source with a compatible license) an implementation for MFCCs. This implementation should be able to take an [login to view URL] as an input and transform it to a double[] vector where each element of this double vector comes from the MFCC computation;

2) Integrating the implementation into the [login to view URL] Framework;

3) Providing unit tests that pass;

4) Writing an example using those MFCCs to perform sound classification using any of the standard machine learning methods already present in the framework (i.e. SVMs). The examples do not have to use real data or tackle any specific problem, just show how they could be used with a machine learning method. More details about this will be given below. The example should be able to read a collection of .wav files (e.g. [login to view URL], or your own collection of sounds) and perform classification using any machine learning method such as SVMs.

5) Providing those implementations and unit tests as a pull-request to the project's issue tracker ( [login to view URL] ). All bidders can also list their questions in Issue 441 ( [login to view URL] ) which is aimed to track development of this issue.

Note 1: The proposed project should not require any large changes in the framework outside of the [login to view URL] namespace. Once a method for computing MFCCs is implemented, the framework should already be able to provide all the rest of the machinery for building a machine learning method using those MFCCs. Examples on how to create a machine learning method for classification are shown, for example, in the bottom page of [login to view URL]

Note 2: The employer can help with any of the steps described above, with special emphasis to item 4.

Note 3: A bonus of 10% over the bid value will be granted for blogging about the implementation or the use of the implementation fo training machine learning algorithms. This should be an easy addition after the last step specified above (writing an example) is also completed.

Note 4: The implementation should be made available under the MIT or BSD licenses. This means that it is perfectly OK to re-use existing open-source implementations for MFCCs as long as they are also available under the MIT or BSD licenses. Good examples of existing implementations that you would be free to port (and copy & paste as you'd like) are [login to view URL] and [login to view URL]://[login to view URL] .

Before bidding, please make sure that you are able to clone the [login to view URL] Framework project repository and build the project.

The description of this project might be updated depending on questions asked by bidders. Please also keep track of the issue at [login to view URL] for additional clarifications.

.NET C# Programming Machine Learning (ML) Software Architecture

Project ID: #14978191

About the project