Spectral Analysis March 3, 2016 Mini-Rant I have succeeded in grading your course project reports. Things to keep in mind: A table of stop phonemes is.

Slides:



Advertisements
Similar presentations
Acoustic/Prosodic Features
Advertisements

Harmonics October 29, 2012 Where Were We? Were halfway through grading the mid-terms. For the next two weeks: more acoustics Its going to get worse before.
Spectral Analysis Feburary 24, 2009 Sorting Things Out 1.TOBI transcription homework rehash. And some structural reminders. 2.On Thursday: back in the.
Frequency Domain The frequency domain
Physics 1251 The Science and Technology of Musical Sound Unit 1 Session 8 Harmonic Series Unit 1 Session 8 Harmonic Series.
DFT/FFT and Wavelets ● Additive Synthesis demonstration (wave addition) ● Standard Definitions ● Computing the DFT and FFT ● Sine and cosine wave multiplication.
Intro to Spectral Analysis and Matlab. Time domain Seismogram - particle position over time Time Amplitude.
Fourier Transform – Chapter 13. Image space Cameras (regardless of wave lengths) create images in the spatial domain Pixels represent features (intensity,
Let’s go back to this problem: We take N samples of a sinusoid (or a complex exponential) and we want to estimate its amplitude and frequency by the FFT.
Intro to Spectral Analysis and Matlab Q: How Could you quantify how much lower the tone of a race car is after it passes you compared to as it is coming.
Signals Processing Second Meeting. Fourier's theorem: Analysis Fourier analysis is the process of analyzing periodic non-sinusoidal waveforms in order.
Fourier Analysis D. Gordon E. Robertson, PhD, FCSB School of Human Kinetics University of Ottawa.
Time and Frequency Representation
Unit 7 Fourier, DFT, and FFT 1. Time and Frequency Representation The most common representation of signals and waveforms is in the time domain Most signal.
Basics of Signal Processing. frequency = 1/T  speed of sound × T, where T is a period sine wave period (frequency) amplitude phase.
Representing Acoustic Information
Source/Filter Theory and Vowels February 4, 2010.
Sampling Theorem, frequency resolution & Aliasing The Sampling Theorem will be the single most important constraint you'll learn in computer-aided instrumentation.
LE 460 L Acoustics and Experimental Phonetics L-13
Where we’re going Speed, Storage Issues Frequency Space.
Basics of Signal Processing. SIGNALSOURCE RECEIVER describe waves in terms of their significant features understand the way the waves originate effect.
Motivation Music as a combination of sounds at different frequencies
Automatic Pitch Tracking September 18, 2014 The Digitization of Pitch The blue line represents the fundamental frequency (F0) of the speaker’s voice.
Resonance, Revisited March 4, 2013 Leading Off… Project report #3 is due! Course Project #4 guidelines to hand out. Today: Resonance Before we get into.
FOURIER SERIES §Jean-Baptiste Fourier (France, ) proved that almost any period function can be represented as the sum of sinusoids with integrally.
Acoustic Analysis of Speech Robert A. Prosek, Ph.D. CSD 301 Robert A. Prosek, Ph.D. CSD 301.
Vowel Acoustics November 2, 2012 Some Announcements Mid-terms will be back on Monday… Today: more resonance + the acoustics of vowels Also on Monday:
Signals & Systems Lecture 11: Chapter 3 Spectrum Representation (Book: Signal Processing First)
Fourier series. The frequency domain It is sometimes preferable to work in the frequency domain rather than time –Some mathematical operations are easier.
Transforms. 5*sin (2  4t) Amplitude = 5 Frequency = 4 Hz seconds A sine wave.
Spectral Analysis AOE March 2011 Lowe 1. Announcements Lectures on both Monday, March 28 th, and Wednesday, March 30 th. – Fracture Testing –
Vibrate the membrane Acoustic wave.
Harmonics November 1, 2010 What’s next? We’re halfway through grading the mid-terms. For the next two weeks: more acoustics It’s going to get worse before.
1 LES of Turbulent Flows: Lecture 2 Supplement (ME EN ) Prof. Rob Stoll Department of Mechanical Engineering University of Utah Fall 2014.
Seismic Reflection Data Processing and Interpretation A Workshop in Cairo 28 Oct. – 9 Nov Cairo University, Egypt Dr. Sherif Mohamed Hanafy Lecturer.
Formants, Resonance, and Deriving Schwa March 10, 2009.
Resonance October 23, 2014 Leading Off… Don’t forget: Korean stops homework is due on Tuesday! Also new: mystery spectrograms! Today: Resonance Before.
Fourier series: Eigenfunction Approach
Pre-Class Music Paul Lansky Six Fantasies on a Poem by Thomas Campion.
Voice Quality + Korean Stops October 16, 2014 Don’t Forget! The mid-term is on Tuesday! So I have a review sheet for you. For the mid-term, we will just.
Vowel Acoustics March 10, 2014 Some Announcements Today and Wednesday: more resonance + the acoustics of vowels On Friday: identifying vowels from spectrograms.
421 Pendulum Lab (5pt) Equation 1 Conclusions: We concluded that we have an numerically accurate model to describe the period of a pendulum at all angles.
Spectral Analysis Feburary 23, 2010 Sorting Things Out 1.On Thursday: back in the computer lab. Craigie Hall D 428 Analysis of Korean stops. 2.Remember:
Frequency, Pitch, Tone and Length February 12, 2014 Thanks to Chilin Shih for making some of these lecture materials available.
7- 1 Chapter 7: Fourier Analysis Fourier analysis = Series + Transform ◎ Fourier Series -- A periodic (T) function f(x) can be written as the sum of sines.
Resonance January 28, 2010 Last Time We discussed the difference between sine waves and complex waves. Complex waves can always be understood as combinations.
Resonance March 7, 2014 Looking Ahead I’m still behind on grading the mid-term and Production Exercise #1… They should be back to you by Friday.
The Spectrum n Jean Baptiste Fourier ( ) discovered a fundamental tenet of wave theory.
The Frequency Domain Digital Image Processing – Chapter 8.
Spectral analysis and discrete Fourier transform Honza Černocký, ÚPGM.
Basic Acoustics + Digital Signal Processing January 11, 2013.
Resonance October 29, 2015 Looking Ahead I’m still behind on grading the mid-term and Production Exercise #1… They should be back to you by Monday. Today:
Harmonics October 28, Where Were We? Mid-terms: our goal is to get them back to you by Friday. Production Exercise #2 results should be sent to.
FOURIER THEORY: KEY CONCEPTS IN 2D & 3D
CS 591 S1 – Computational Audio
Spectrum Analysis and Processing
CS 591 S1 – Computational Audio
Spectral Analysis Spectral analysis is concerned with the determination of the energy or power spectrum of a continuous-time signal It is assumed that.
MECH 373 Instrumentation and Measurements
Continuous-Time Signal Analysis
Intro to Fourier Series
Image Processing, Leture #14
FFTs, Windows, and Circularity
Lab 6: Sound Analysis Fourier Synthesis Fourier Analysis
Sound shadow effect Depends on the size of the obstructing object and the wavelength of the sound. If comparable: Then sound shadow occurs. I:\users\mnshriv\3032.
7.2 Even and Odd Fourier Transforms phase of signal frequencies
INTRODUCTION TO THE SHORT-TIME FOURIER TRANSFORM (STFT)
Discrete Fourier Transform
The Frequency Domain Any wave shape can be approximated by a sum of periodic (such as sine and cosine) functions. a--amplitude of waveform f-- frequency.
Geol 491: Spectral Analysis
Presentation transcript:

Spectral Analysis March 3, 2016

Mini-Rant I have succeeded in grading your course project reports. Things to keep in mind: A table of stop phonemes is a theory about how your language works. Your job is to back up that theory with evidence. Minimal pairs are not that important, but examples are. You are a better linguist than your consultant is. You cannot expect your consultant to know what a “minimal pair” is, or to be able to produce them on demand for you. Your consultant may very well say things about their language that are completely uninformed and wrong.

Mini-Rant, part II Do not tell me about the writing system for your language. Your job is not to find out how the language is written, but how it is spoken. Be conservative with your conclusions. Only tell me what can be justified with evidence from the data you’ve collected. The process is more important than the product. I would rather that you get comfortable with not knowing what the “right” answer is than for you to tell me what somebody else thinks the right answer is.

Moving On… Don’t forget: the next course project report is due on Tuesday! On Tuesday, we will also start adding mystery spectrograms into the mix. …and I will return your mid-terms to you. Today’s goal: let’s learn where spectrograms come from.

The Source The complex wave emitted from the glottis during voicing= The source of all voiced speech sounds. In speech (particularly in vowels), humans can shape this spectrum to make distinctive sounds. Some harmonics may be emphasized... Others may be diminished (damped) Different spectral shapes may be formed by particular articulatory configurations....but the process of spectral shaping requires the raw stuff of the source to work with.

Spectral Shaping Examples Certain spectral shapes seem to have particular vowel qualities.

Spectrograms A spectrogram represents: Time on the x-axis Frequency on the y-axis Intensity on the z-axis

Real Vowels

Ch-ch-ch-ch-changes Check out some spectrograms of sinewaves which change frequency over time:

The Whole Thing What happens when we put all three together? This is an example of sinewave speech.

The Real Thing Spectral change over time is the defining characteristic of speech sounds.  It is crucial to understand spectrographic representations for the acoustic analysis of speech.

Life’s Persistent Questions How do we get from here: To here? Answer: Fourier Analysis

Fourier’s Theorem Joseph Fourier ( ) French mathematician Studied heat and periodic motion His idea: any complex periodic wave can be constructed out of a combination of different sinewaves. The sinusoidal (sinewave) components of a complex periodic wave = harmonics

Fourier Analysis Building up a complex wave from sinewave components is straightforward… Breaking down a complex wave into its spectral shape is a little more complicated. In our particular case, we will look at: Discrete Fourier Transform (DFT) Also: Fast Fourier Transform (FFT) is used often in speech analysis Basically a more efficient, less accurate method of DFT for computers.

Spectral Slices The first step in Fourier Analysis is to window the signal. I.e., break it all up into a series of smaller, analyzable chunks. This is important because the spectral qualities of the signal change over time. a “window” Check out the typical window length in Praat.

The Basic Idea For the complex wave extracted from each window... Fourier Analysis determines the frequency and intensity of the sinewave components of that wave. Do this about 1000 times a second, turn the spectra on their sides, and you get a spectrogram.

Possible Problems What would happen if a waveform chunk was windowed like this? Remember, the goal is to determine the frequency and intensity of the sinewave components which make up that slice of the complex wave.

The Usual Solution The amplitude of the waveform at the edges of the window is normally reduced... by transforming the complex wave with a smoothing function before spectral analysis. Each function defines a particular window type. For example: the “Hanning” Window

There are lots of different window types... each with its own characteristic shape Hamming BartlettGaussian HanningWelchRectangular

Window Type Ramifications Play around with the different window types in Praat.

Ideas Once the waveform has been windowed, it can be boiled down into its component frequencies. Basic strategy: Determine whether the complex wave correlates with sine (and cosine!) waves of particular frequencies. Correlation measure: “dot product” = sum of the point-by-point products between waves. Interesting fact: Non-zero correlations only emerge between the complex wave and its harmonics! (This is Fourier’s great insight.)

A Not-So-Complex Example Let’s build up a complex wave from 8 samples of a 1 Hz sine wave and a 4 Hz cosine wave. Note: our sample rate is 8 Hz A1 Hz B4 Hz CSum: Check out a visualization.

Correlations, part 1 Let’s check the correlation between that wave and the 1 Hz sinewave component CSum: A1 Hz: C*ADot: The sum of the products of each sample is 4. This also happens to be the dot product of the 1 Hz wave with itself. = its “power”

Correlations, part 2 Let’s check the correlation between the complex wave and a 2 Hz sinewave (a non-component) CSum: D2 Hz: C*DDot: The sum of the products of each sample is 0.  We now know that 2 Hz was not a component frequency of the complex wave.

Correlations, part 3 Last but not least, let’s check the correlation between the complex wave and the 4 Hz cosine wave CSum: B4 Hz C*BDot: The sum of the products of each sample is 8. Yes, 8 happens to be the dot product of the 4 Hz wave with itself. its “power”

Mopping Up Our component analysis gave us the following dot products: C*A = 4(A = 1 Hz sinewave) C*D = 0(D = 2 Hz sinewave) C*B = 8(B = 4 Hz cosine wave) We have to “normalize” these products by dividing them by the power of the “reference” waves: power (A) = A*A = 4  C*A/A*A = 4/4 = 1 power (D) = D*D = 4  C*D/D*D = 0/4 = 0 power (B) = B*B = 8  C*B/B*B = 8/8 = 1 These ratios are the amplitudes of the component waves.

Let’s Try Another Let’s construct another example: 1 Hz sinewave + a 4 Hz cosine wave with half the amplitude A1 Hz *B4 Hz ESum: Let’s check the 1 Hz wave first: ESum: A1 Hz E*ADot: Sum = 4

Yet More Dots Another example: 1 Hz sinewave + a 4 Hz cosine wave with half the amplitude. Now let’s check the 4 Hz wave: ESum: B4 Hz E*BDot: The sum of these products is also 4. = half of the power of the 4 Hz cosine wave.  The 4 Hz component has half the amplitude of the 4 Hz cosine reference wave. (we know the reference wave has amplitude 1)

Mopping Up, Part 2 Our component analysis gave us the following dot products: E*A = 4(A = 1 Hz sinewave) E*B = 4(B = 4 Hz cosine wave) Let’s once again normalize these products by dividing them by the power of the “reference” waves: power (A) = A*A = 4  E*A/A*A = 4/4 = 1 power (B) = B*B = 8  E*B/B*B = 4/8 =.5 These ratios are the amplitudes of the component waves. The 1 Hz sinewave component has amplitude 1 The 4 Hz cosine wave component has amplitude.5

Footnote Sinewaves and cosine waves are orthogonal to each other.  The dot product of a sinewave and a cosine wave of the same frequency is Asin Fcos A*FDot: However, adding cosine and sine waves together simply shifts the phase of the complex wave. Check out different combos in Praat.

Problem #1 For any given window, we don’t know what the phase shift of each frequency component will be. Solution: 1.Calculate the correlation with the sinewave 2.Calculate the correlation with the cosine wave 3.Combine the resulting amplitudes with the pythagorean theorem: Take a look at the java applet online:

Sine + Cosine Example Let’s add a 1 Hz cosine wave, of amplitude.5, to our previous combination of 1 Hz sine and 4 Hz cosine waves C1+4: *Fcos GSum: Let’s check the 1 Hz sine wave again: GSum: A1 Hz G*ADot: Sum = 4

Sine + Cosine Example Now check the 1 Hz cosine wave: GSum: F1 Hz G*FDot: Sum = 2 Sinewave component amplitude = 4/4 = 1 Cosine wave component amplitude = 2/4 =.5 Total amplitude = Check out the amplitude of the combo in Praat.

In Sum To perform a Fourier analysis on each (smoothed) chunk of the waveform: 1.Determine the components of each chunk using the dot product-- Components yield a dot product that is not 0 Non-components yield a dot product that is 0 2.Normalize the amplitude values of the components  Divide the dot products by the power of the reference wave at that frequency 3.If there are both sine and cosine wave components at a particular frequency: Combine their amplitudes using the Pythagorean theorem