16

By long shot it is doable - to what extend? You will see. This task of environmental sound classification is not very well studied. Also choice of machine learning paradigm is crucial - statistical approach or maybe binary classifier? You can start with GMM's, ANN's and SVM's - I opt for GMM's and ANN's. Yes, most of people are using MFCC's because they are ...


11

Some Features: Mean. Variance. Skewness. Kurtosis. Dominant 3 frequencies in the DFT. Energy of the 3 dominant frequencies. Max Value. Min Value. Median. Total Variation. Usually I'd compute them in running windows. Another great information is the Histogram of the Derivative. Or just all the above of the Derivative.


6

To complete bjou's answer: A potential problem with min-max normalization is that it is very sensitive to outliers. For example, if your dataset contain 100 two-dimensional examples; and that the dataset looks like this: 1, 147 8, -252 3, 125 2, -605 ... 10000, -100 <- outlier 4, 200 min-max normalization will squash almost all values of the first ...


6

Both are reasonable approaches and it is foreseeable that either one could outperform the other empirically. The Euclidean distance assumes the data to be isotropically Gaussian, i.e. it will treat each feature equally. On the other hand, the Mahalanobis distance seeks to measure the correlation between variables and relaxes the assumption of the Euclidean ...


6

Freesound is a repository of sound files categorised by user-defined tags. Through those, you can spot what other users have labeled industrial sounds but be prepared that some of those might be artificially created (so, not really a recording of an engine but something that sounds like an engine). In the same category is also Soundsnap. The other popular ...


5

In addition to the features mentioned so far I would like to mention measures of complexity such as: Shannon Entropy LZ Complexity Fractal Dimension There are also Fourier Descriptors (as hinted by Drazick already) and their equivalent in Wavelet Analysis and of course simple histogram bins which would return how frequently each gear is engaged en route. ...


5

I will try to give you some intuition into it by a different example. Think we have 3 machines which can generate the numbers 1, 2, 3. The first machine generates the number 1 with 80% and the numbers 2, 3 with 10% each. The second machine generates the number 2 with 80% and the numbers 1, 3 with 10% each. The third machine generates the number 3 with 80% ...


5

You are right that a distributed system could be "something like a transmission line". Note that the system $$y(t)=x(t-T)\tag{1}$$ is a simple model of a transmission line, where just a frequency-independent delay $T$ is taken into account, and the attenuation is neglected. Note that lumped electrical systems, described by resistors, capacitors and ...


4

Non-verbal Audio (let alone environmental) seems to be the little brother to main stream machine learning media types like images, speech, text. To answer your question is it possible to train a network to identify a given sound? Yes, it is! But it's hard for all the same reasons machine learning is hard. However what's really holding Audio back, and why ...


4

There is a package called FukuML. In their description (Version 0.4.1) they write: Support Vector Machine Primal Hard Margin Support Vector Machine Binary Classification Learning Algorithm Dual Hard Margin Support Vector Machine Binary Classification Learning Algorithm Polynomial Kernel Support Vector Machine Binary Classification Learning Algorithm ...


3

Here is a solution for sound classification for 10 classes: dog barking, car horn, children playing etc. It is based on tensorflow library using neural networks. Features are extracted by converting sound clips to spectrogram


3

I would start with following set of parameters: MFCC's (I that know you tried it, but stay with me) without static energy (1'st coefficient) Some descriptors from MPEG-7, like: Spectral Flatness, Harmonicity, Fundamental frequency, Spectral Spread, etc. You can find out more about them here: click or in this great book: MPEG-7 Audio and Beyond You should ...


3

Labelling a missing value as -1000 (assuming a non-missing value is much smaller in absolute value) will not cause problems with anything based on decision trees/stumps (bagged/boosted). You might also have some success with mixture models, for example a GMM - the learning algorithm will allocate some components to cover the missing data. Same for non-...


3

As Conrad pointed out, a correlator is probably your best bet. The correlation of a signal with itself (also known as its self-similarity) is larger than its correlation with any other signal (except for a constant factor related to the signals' energy). In your case, you would implement two correlators, one for Signal 1 and one for Signal 2. Then, you'd ...


3

Hello I will be brief and I hope you understand, due to the shape of your signal I think it is best treated with wavelet transform base HAAR, the reason for using this transform is that it will give a representation in time and frequency where you can get the relevant information of the signal, now an important parameter is that you use the base HAAR (there ...


3

You should use something like Robust Convex Clustering: Assume that we are given $ n $ data points and each data point is described by a $ d $ dimensional feature vector. We collectively represent the data by a data matrix $ X \in \mathbb{R}^{d \times n} $. The convex clustering method clusters data points into groups via solving a convex optimization ...


3

Have a look at these two python libraries that provide a number of audio features easily from WAV files, including MFCC. Librosa: MFCC docs, github Madmom: MFCC docs, github Good luck!


3

You should use the one you need for your problem, when you know which components of your signal are of interest to you. Let's say you have in your electronic editing an ADC digitizing 40M samples per second to study a heart rate of 70 beats per minute, you are very likely to work with useless information, that's why it will be better to down-sample your ...


3

From what you've mentioned it looks like the task is for environmental sound event detection. I think that the best starting point for you is to check the DCASE challenge (Detection and Classification of Acoustic Scenes and Events). The result pages are amazing - you can sort all systems by their performance, classifier being used, features, etc. For example ...


2

Referring to MATLAB, the basic steps are Determine the connected components: CC = bwconncomp(BW, conn); Compute the area of each component: S = regionprops(CC, 'Area'); Remove small objects: L = labelmatrix(CC); BW2 = ismember(L, find([S.Area] >= P)); At the last step, after obtaining $L$, you might as well retain the component with the largest area ...


2

Looks like you are using Matlab. Try bwareaopen(I, N), where I is the original binary image and N is the estimated size of each unwanted connected region. You can try edit bwareaopen for more details. Basically the algorithm tries to find the size of connected regions. Connected-component labeling with union-find algorithm is expected to get you there.


2

This article deals exactly with the same type of supervised classification based on labelled GLCM classes: GLCM Textural Features for Brain Tumor Classification


2

From one perspective, not much. However, biological perceptual systems have evolved in response to physical properties of natural environments. And such environments possess certain characteristics, such as Laplacian distribution shaped histograms. Therefore natural images are grouped and analyzed together.


2

No, that's not it. Look at the last sentence of your quote: ... which may be enforced by scaling the data to some precision and truncating to integer values. So this is exactly what you do. Take all your feature vectors, multiply them by 1000 (or some other factor), and truncate to integer. Then the distance between any two unique feature vectors will ...


2

The edges of the shadows are crisper where they are nearer to the shadow caster. Also, since the sun is so far away, beams of light are essentially parallel when they reach us, which means that shadows from sunlight are basically orthographic projections of the shadow casters. Also, if you knew an objects dimensions and saw its shadow on something that you ...


2

Shadow has very specific properties that makes it very clear way of making it distinguishable from the regular object. A lot of work in the area of background subtraction and surveillance has been using this to eliminate the shadows or to avoid them being mistaken as the actual object (or human). As observed by Daniel Grest To distinguish the shadows ...


2

The characteristics of the shadow are as follows: It is always dark regardless of the color of the object or the color of the light used to make a shadow It only shows a dark outline of the object It is formed in the opposite direction to the source of light The size of the shadow depends upon the distance between the source of light and the opaque object ...


2

I do not think that there is a standard approach to this problem, because as far as I know it has not yet been solved satisfactorily. I believe that you can find a solution to a (very) simplified version of the problem by restricting it for example to the recognition of a fixed set of animal sounds. The way to go would probably be to use methods from speech ...


2

Following my comment, I now have time to post it as an answer. If you expect that a large enough portion of the vectors belong to that line, you can use RANSAC. RANSAC works by selecting a small and random subset of the data and fit it to the desired model (say a line). The fitting can be achieved in any way you like, e.g. least-squares. Then, the data is ...


Only top voted, non community-wiki answers of a minimum length are eligible