Signal Feature Extractor

Extract and analyze signal features

Since R2025a

Description

Use Signal Feature Extractor to extract time-domain and frequency-domain features from a signal. You can use the extracted features to train a machine learning model or a deep learning network. Using the app, you can:

• Extract statistical, pulse metrics, and harmonic time-domain features.

• Extract statistical, bandwidth, and peak-based frequency-domain features.

• Extract audio features from imported audio file data. (since R2026a)

• Use entire signals or frames.

• Use parallel computation to accelerate feature extraction.

• Rank signal features using feature-ranking algorithms for classification workflows.

• Export labeled signal sets and extracted features.

Signal Feature Extractor saves data as labeledSignalSet objects, MATLAB^® tables, or matrices. You can use labeledSignalSet objects to train a deep learning classifier or analyze data and report statistics.

For more information, see Use Signal Feature Extractor App.

Open the Signal Feature Extractor App

• MATLAB Toolstrip: On the Apps tab, under Signal Processing and Audio, click the app icon.

• MATLAB command prompt: Enter signalFeatureExtractor.

Examples

Extract Signal Features from Unlabeled Whale Songs

Extract the RMS, crest factor, and impulse factor features from a multichannel signal.

Load the data set labelwhalesignals that includes two recordings (whale1 and whale2) of blue whale songs of trills and moans at a sample rate Fs. For more information about the whale songs data set, see Label Signal Attributes, Regions of Interest, and Points.

load labelwhalesignals

Define a two-channel signal x as the amplitude of the whales' songs between 2.3819 seconds and 3.2175 seconds. The sample rate Fs is 4 kHz. At this time interval, the song of whale1 is a trill and that of whale2 is a moan.

songIdx = round(2.3819*Fs):round(3.2175*Fs);
x = [whale1(songIdx) whale2(songIdx)];
strips(x)
yticklabels(["Whale2 Moan" "Whale1 Trill"])

Extract frame-based signal features in the time and frequency domain.

1. Open Signal Feature Extractor. On the Extract Features tab, click New and select From Workspace in the Members list.

2. In the dialog box, select the signal x. Add time information: Select Time from the drop-down list and enter the sample rate Fs, which is measured in Hz.

3. Click Import and Close to import the signal and then close the dialog box. The signal appears in the Members browser.

4. To extract features per signal frame, select Frame-based in the Frame Policy section. Specify a frame size and frame rate of 64 milliseconds and 50 milliseconds, respectively.

5. Add time-domain features: Click Time-Domain in the Feature Generation gallery. Select RMS, Crest factor, and Impulse Factor. Click Add Features.

6. Add frequency-domain features: Click Frequency-Domain in the Feature Generation gallery. Select Mean Frequency, and Welch's PSD. Click Add Features.

7. To extract the selected signal features, click Extract . The app shows the extracted features in the table FeatureTable1. To see the features along the frames, expand the arrow under the Name column. Each column contains a feature. Welch's PSD vector feature is listed in several columns.

Programmatic Use

signalFeatureExtractor opens the Signal Feature Extractor app.

Tips

If you have Parallel Computing Toolbox™, you can run multiple feature extraction jobs at the same time. Click Use Parallel to enable the parallelization of the feature extraction. For more information, see Accelerate Feature Extraction with Parallel Computing.

Version History

Introduced in R2025a

R2026a: Extract features from audio signals

Extract features from audio signals using the Signal Feature Extractor app. You must have a license for Audio Toolbox™ to use this functionality.

• Import audio signals from audio files or labeled signal sets pointing to audioDatastore (Audio Toolbox) objects.

• Extract time-domain, frequency-domain, and audio features from audio signals for use in AI models. Audio features include mel spectrogram, mel-frequency cepstral coefficients (MFCC), pitch, and spectral centroid.