Spectrogram of the spoken words "nineteenth century". Frequencies are shown increasing up the vertical axis, and time on the horizontal axis. The legend to the right shows that the color intensity increases with the density.A 3D spectrogram: The RF spectrum of a battery charger is shown over time
Aspectrogram is a visual representation of thespectrum offrequencies of a signal as it varies with time. When applied to anaudio signal, spectrograms are sometimes calledsonographs,voiceprints, orvoicegrams. When the data are represented in a 3D plot they may be calledwaterfall displays.
A common format is a graph with two geometric dimensions: one axis representstime, and the other axis representsfrequency; a third dimension indicating theamplitude of a particular frequency at a particular time is represented by theintensity or color of each point in the image.
There are many variations of format: sometimes the vertical and horizontal axes are switched, so time runs up and down; sometimes as awaterfall plot where the amplitude is represented by height of a 3D surface instead of color or intensity. The frequency and amplitude axes can be eitherlinear orlogarithmic, depending on what the graph is being used for. Audio would usually be represented with a logarithmic amplitude axis (probably indecibels, or dB), and frequency would be linear to emphasize harmonic relationships, or logarithmic to emphasize musical, tonal relationships.
3D surface spectrogram of a part from a music piece.
Spectrogram of a male voice saying 'ta ta ta'.
Spectrogram of dolphin vocalizations; chirps, clicks and harmonizing are visible as inverted Vs, vertical lines and horizontal striations respectively.
Spectrogram of anFM signal. In this case the signalfrequency is modulated with asinusoidal frequency vs. time profile.
Spectrum above and waterfall (Spectrogram) below of an 8MHz widePAL-I Television signal.
Spectrograms of light may be created directly using anoptical spectrometer over time.
Spectrograms may be created from atime-domain signal in one of two ways: approximated as a filterbank that results from a series ofband-pass filters (this was the only way before the advent of modern digital signal processing), or calculated from the time signal using theFourier transform. These two methods actually form two differenttime–frequency representations, but are equivalent under some conditions.
The bandpass filters method usually usesanalog processing to divide the input signal into frequency bands; the magnitude of each filter's output controls a transducer that records the spectrogram as an image on paper.[3]
Creating a spectrogram using the FFT is adigital process. Digitallysampled data, in thetime domain, is broken up into chunks, which usually overlap, and Fourier transformed to calculate the magnitude of the frequency spectrum for each chunk. Each chunk then corresponds to a vertical line in the image; a measurement of magnitude versus frequency for a specific moment in time (the midpoint of the chunk). These spectrums or time plots are then "laid side by side" to form the image or a three-dimensional surface,[4] or slightly overlapped in various ways, i.e.windowing. This process essentially corresponds to computing the squaredmagnitude of theshort-time Fourier transform (STFT) of the signal — that is, for a window width,.[5]
From the formula above, it appears that a spectrogram contains no information about the exact, or even approximate,phase of the signal that it represents. For this reason, it is not possible to reverse the process and generate a copy of the original signal from a spectrogram, though in situations where the exact initial phase is unimportant it may be possible to generate a useful approximation of the original signal. The Analysis & Resynthesis Sound Spectrograph[6] is an example of a computer program that attempts to do this. Thepattern playback was an early speech synthesizer, designed atHaskins Laboratories in the late 1940s, that converted pictures of the acoustic patterns of speech (spectrograms) back into sound.
In fact, there is some phase information in the spectrogram, but it appears in another form, as time delay (orgroup delay) which is thedual of theinstantaneous frequency.[7]
The size and shape of the analysis window can be varied. A smaller (shorter) window will produce more accurate results in timing, at the expense of precision of frequency representation. A larger (longer) window will provide a more precise frequency representation, at the expense of precision in timing representation. This is an instance of theHeisenberg uncertainty principle, that the product of the precision in twoconjugate variables is greater than or equal to a constant (B*T>=1 in the usual notation).[8]
Early analog spectrograms were applied to a wide range of areas including the study of bird calls (such as that of thegreat tit), with current research continuing using modern digital equipment[9] and applied to all animal sounds. Contemporary use of the digital spectrogram is especially useful for studyingfrequency modulation (FM) in animal calls. Specifically, the distinguishing characteristics of FM chirps, broadbandclicks, and social harmonizing are most easily visualized with the spectrogram.
Spectrograms are useful in assisting in overcoming speech deficits and in speech training for the portion of the population that is profoundlydeaf.[10]
In deep learning-keyed speech synthesis, spectrogram (or spectrogram inmel scale) is first predicted by a seq2seq model, then the spectrogram is fed to a neural vocoder to derive the synthesized raw waveform.
By reversing the process of producing a spectrogram, it is possible to create a signal whose spectrogram is an arbitrary image. This technique can be used to hide a picture in a piece of audio and has been employed by severalelectronic music artists.[13] See alsoSteganography.
Some modern music is created using spectrograms as an intermediate medium; changing the intensity of different frequencies over time, or even creating new ones, by drawing them and then inverse transforming. SeeAudio timescale-pitch modification andPhase vocoder.
Spectrograms can be used to analyze the results of passing a test signal through a signal processor such as a filter in order to check its performance.[14]
High definition spectrograms are used in the development of RF and microwave systems.[15]
Spectrograms are now used to displayscattering parameters measured with vector network analyzers.[16]
For a vibration signal, a spectrogram's color scale identifies the frequencies of a waveform's amplitude peaks over time. Unlike a time or frequency graph, a spectrogram correlates peak values to time and frequency. Vibration test engineers use spectrograms to analyze the frequency content of a continuous waveform, locating strong signals and determining how the vibration behavior changes over time.[22]
Spectrograms can be used to analyze speech in two different applications: automatic detection of speech deficits in cochlear implant users and phoneme class recognition to extract phone-attribute features.[23]
In order to obtain a speaker's pronunciation characteristics, some researchers proposed a method based on an idea from bionics, which uses spectrogram statistics to achieve a characteristic spectrogram to give a stable representation of the speaker's pronunciation from a linear superposition of short-time spectrograms.[24]
Researchers explore a novel approach to ECG signal analysis by leveraging spectrogram techniques, possibly for enhanced visualization and understanding. The integration of MFCC for feature extraction suggests a cross-disciplinary application, borrowing methods from audio processing to extract relevant information from biomedical signals.[25]
Accurate interpretation of temperature indicating paint (TIP) is of great importance in aviation and other industrial applications. 2D spectrogram of TIP can be used in temperature interpretation.[26]
The spectrogram can be used to process the signal for the rate of change of the human thorax. By visualizing respiratory signals using a spectrogram, the researchers have proposed an approach to the classification of respiration states based on a neural network model.[27]
^Boashash, B. (1992). "Estimating and interpreting the instantaneous frequency of a signal. I. Fundamentals".Proceedings of the IEEE.80 (4). Institute of Electrical and Electronics Engineers (IEEE):520–538.doi:10.1109/5.135376.ISSN0018-9219.
^Saunders, Frank A.; Hill, William A.; Franklin, Barbara (1 December 1981). "A wearable tactile sensory aid for profoundly deaf children".Journal of Medical Systems.5 (4):265–270.doi:10.1007/BF02222144.PMID7320662.S2CID26620843.
