Speaker recognition systems, auditory analyais is a eru dppreacneg.° UatON Analysis in speaker recognition are the spectipraphic approaches, hing individuals based on their vocal characteristics, T Broach and voles print ident \ocomn ification. Lets explore each of the pectrorraphic Approach: Principle: The spectrographie approach involves the anal N generates a visual representation known represented by color of shading, Features: isis of the frequency content of speech sign Extracts features such as formants, piteh, and other 5} Application: ectral characteristics from the spectrogram, Commonly used in forensic voice analysis and speech signal processing. * Provides a visual tool for the examination of speech signals, ice Print Identification: Principle: + Voice print identification focuses ‘on extracting and analyzing unique characteristics of an individual's voi vocal signature." 8 a catinetie “voice print” oF + Hinvolves features related to the individuals voral tract. piteh, intonation. and other veice charactensstics, | Features: * Extracts features related to the shape and size of the vocal tract. pitch patterns, speech rate, and ether personalized attributes Application: + Used for speaker recognition. including both speaker verification known voices) + Applied in secunty, access control. and law enforcement {confirming identity) and speaker identifeation finding a match in database ot m-ECOGNITION SYSTEM OF APHICComparison: Nature of Analysis: * Spectrographic Approach: Primarily analyzes the frequency content of speech signals, providing a visual representation. * Voice Print Identification: Focuses on extracting specific features related to individual voice characteristics. Applications: * Spectrographic Approach: Often used in forensic analysis and visual inspection of speech signals. * Voice Print Identification: Applied in speaker recognition systems for authentication and identification purposes. Representation: * Spectrographic Approach: Provides a visual representation of the speech signal. * Voice Print Identification: Focuses on creating @ mathematical representation or model of the unique voice characteristics. Use Cases: * Spectrographic Approach: Useful for analyzing speech signals in a more general context, not necessarily for individual speaker identification. * Voice Print Identification: Specifically designed for speaker recognition tasks, including verification and identification.Comparison: Nature of Analysis: * Spectrographic Approach: Primarily analyzes the frequency content of speech signals, providing a visual representation. * Voice Print Identification: Focuses on extracting specific features related to individual voice characteristics. Applications: * Spectrographic Approach: Often used in forensic analysis and visual inspection of speech signals. * Voice Print Identification: Applied in speaker recognition systems for authentication and identification purposes. Representation: * Spectrographic Approach: Provides a visual representation of the speech signal. * Voice Print Identification: Focuses on creating a mathematical representation or model of the unique voice characteristics. Use Cases: * Spectrographic Approach: Useful for analyzing speech signals in a more general context, not necessarily for individual speaker identification. * Voice Print Identification: Specifically designed for speaker recognition tasks, including verification and identification. OACH ORPrin Spectral Analysis: Spectral analysis involves decomposing the speech signat Into its frequency components MECC mimies the human ear's tesponse to different frequencies by tranaferming the power ‘spectrum of the speech signs Mel-Frequency Cepstral Coefficients (MFCC): Pitch analysis focuses on determining the fundamental frequency (piten) of the speakers Pitch Analysis: Formant analysis identifies and ‘analyzes the resonant frequencies In the vocal tract during speech reduction. FormantAnalysis: Examines characteristics related to the glottal souree of speach, ‘uch a8 the voral fold vibrations, Voice Source Analysis: Prosodic features analyze the rhythm, intonation, and stress ppatterns in speech Prosodic Features: Fourier Transform is commonly used to obtain the spectrum of the ‘speech signal The process involves dividing the ‘speech signal into frames. applyingaa filterbank to obtain mele frequency bins, takingthe fogarthm.and then applying the discrete cosine transform Algorithms such a8 autocorrelation ‘of cepstral analyais are used to estimate piten, Formants are typically identified a5, peaks inthe frequency spectrum. Measures like jitter (frequency Variation) and shimmer (ampitude variation) are used to characterize the voiee source. Exuaction of features related to speach rate, guration of sabes, Venations im pten, and intensity Extracted features may include formants (tesonant frequencies). spectral peaks, and other frequency related characteristics The resulting cootficients capture esvental spectral information and ‘are commonly used in speaker recognition systems, Extracted pitch information can be used to dstingush speakers based ‘on pitch patterns. Formant frequencies and bandwidths ‘provide information about the shape ‘and site of the vocal vact, contributing to speaker eisunctiveness Jitter, shimmer, and other voice source features contnbute to the Uniqueness of a speakers voice. Prosodic features reflect the ‘emotional ang benavioral nepeets of speech. eanancingpeaker Method sol _ Features Recognition system e ; Dynamic Features: Captures the cynamic aspects of De'ta Coefficients represent the Dynamic features enhance the speech by considering changes rate of change of other features, representation cf temporal over time, Providing temporal information. variations in spatch, contributing to the sprahers characterization, Dei ep. earning, Approaches Deep leaming vtlizes neural Convolutional Neural Networks Deep leaming models lean NI NI . ‘Networks to automatically eam (CNNs) or Recurrent Neural a hierarchical representations from Networks (RNNs) may be employed raw speech signals, for feature leaming. data. eliminating the need for handcrafted features. Gaussian Mixture Models GMMs model the probability Trains the model ona setofknown Statistical [characteristics of space * Gstribution of feature vectors for speakers and calculates. Tkelihood features are represented by GWMs (MMs); each speaker using a mixture of ratios during testing. Providing @ model for each Gaussian distributions. speaker, A - UTAis eHfective for capturing LTA); LTAfocuses on capturing longterm Averages features over an. Long Term Averagin statistical information about a extended time window, providing a ‘Speaker characteristics in. speaker's voice. stable representation, ‘continuous speech, izatic |; VOrepresents speech features by Quantizes feature vectors intoa = VQis efficient for storage and Vector Quantization (v mapping themto asset of limited set of codewords. reducing comparison of speech patterns, tepresentative vectors (codebook). dimensionality, often used in conjunction with, ‘other methods, 7 HMMs model the temporal Trains HMIMs on speech dataand —HMthis are effective for capturing Hidden Markov Models evolution of speech features by uses them to model the dynamics _context-dependent infermation ane representing speech as a of speech, the temporal aspects of speech, sequence of states. RECOGNITION SYST fEMICCaptures the dynamic aspects of Speech by considering changes over time, DeepleamingAppmaches deep teanng tices eva Networks to automaticly eam hierarchical representation from Faw speech signals, Gaussian Mixture Models Mts ‘model the probabiity tribution of feature vectors for ach speaker using a mixture of Gaussian distributions. Long Term Averaging (LTA); LTAfocuses on capturing long-term statistical information abouta speaker's voice, Vector Quantization (VQ); VQrepresents speech features by mapping themtoa set of representative vectors (codebook). Hidden Markov Models HMMs model the temporal (HMMs): iran evolution of speech features by representing speech asa sequence of states. PROACHES 10 SPEAKE! VANDA DP Dea costficients represent the ‘rate of change of ether features, Providing temporal information, Convolutional Neural Netw (CNNs) of Recurrent Neural Networks (RUNS) may bee for feature aming, Trains the model on a set of knoan, ‘speakers and calculates likelihood ‘ratios during testing. Averages features over an extended time window, providing a stable representation. Quantizes feature vectors intoa limited set of codewords, reducing Gimensionalty. Trains HMMs on speech data and uses them to model the dynamics of speech. nt Oynamic features enhance the "epcesetaton ef temporal \ratensin etch, contrbaing tothe speakers charactenzation, Deep leaming models eam cmp patter and ‘representations direct fom the ata, eliminating the need for handoraedfeatres. Statistical characteristics of speech features are represerted by GMM Providing a model for each, speaker, LIAis effective foreapturing speaker charactrsis in continuous speech, VQis efficent for storage and comparison of speech pattems, ten used in conjunction with coer methods, HAIN are effective for capturing ‘context-dependent information and ‘the temporal aspects of speech. -ECOGNITIONISYSTEMOFic Features: Captures tecjnamic aspects ef Detacceficients represent the Dynamic features enhance the Dreniere speeeh byeonseing changes atwotchangeet therfeatres, representation cf terperl overtime, providing temporal infermation, variations in speech, tothe speakers characterization, DeepleamingAppmaches Deep learning vtlizes neural Convolutional Neural Networks Deep learing models leam ‘etiors to automatcalyleam —_(CNNs} ee Recurent Neural complen pattems and Nerarchial representations from Networks (RNS) maybeemplaed —teprecartavers recy trom tbe raw speech signals, forfeatue laming, ata, eliminating te raed for bandoratted features, ian Mixture Models GMs modelthe probably Trains the model ena st ofkngnn Statistica characteristics of speech oe suibuton of feature vectors for speakers andcalou'atesthethoed features are represarted by GMs, ach speaker using a motue of ation during testing. roving a model foreach Gaussian distributions, speaker, LongTerm Averaging(LTA), UAfonses en captuinglngtemm —_Aeagesfeatues ver an UIAis effective for capturing ‘stetiscal information abouta, extended time window, providing a speaker characterctes in speakers voice, Stable representation. Continuoxs speech, Vector Quantization (VQ); VQrepresents speech features by Quantizes feature vectors into a VQis efficient for storage and mapping them to a set of linted set of codewords, reducing Comparison of speech pattems, ‘epresentative vectors (codebook), Gmersionalty, fen used in conjunction with other methods, Hidden Markov Models HMMs modelthe temporal Trans Hs on speech eataand HAIN are fective for capturing HMMs): ‘evolution of speech features by ‘uses them to model the dynamics Contert-dependent information and (HMMs): representing speech asa of speech, eee ‘the temporal aspects cf speech, ‘Sequence . PROACHES TOSPEAKER RECOGNITION'SYSTEMIOF “AUDITORY ANALYSISGAUSSIAN MIXTURE MODELS (GMMS); InitiairedGAUSSIAN MIXTURE MODELS (GMMS):LONG TERM AVERAGING (LTA): Enrollment Phase Testing PhaseAutom: it ated Speaker Principle Cognition system eon syster Dynamic Features: Captures the ¢imamic ESpeCts of SPEECH by con "sidering changes | owines Oeep leaming utiizes neural NetWorhs to automaticalisieam Hierarchical representations fem few speech signals, Gaussian Misture Models che mote! te potabny (GMN fs); Gstribution of feature vectors for ach speaker Using @ mixture of Gaussian distributions, UTA focuses on eapturnglongtem ‘statistical information abouta, Speaker's voice. ‘VQ represents ‘speech features y ‘Mapping them to e set of - HMMs model the temporal tepresenting speech asa saquenea of state, lirted set feos Convolationet Neural Newacis (CNS) er Recurent teu Natwors (a) may b empl for feature lining ‘Trains the model on z set of noxm ‘Speakers and caleulates fcelned ratios during testing, fi cveran ‘tended time nindn, roving stab/2 representation, Quantizes feature vectors into ‘Trains HhIMs on speech data ard cof speech, Naratensin speech cont Nothe speaters characterzaton, Deep learning odes eam comple patems and et rad tor handcrafted features, Statatcal charactersticn of speech features are represented by Git, Proving a modal fer anch speaker, VQ ecient fer storage and are effective fer capturing evel pesch features by uses themtomedel the ¢ramics _cortert dependent information an lution of s : ® the tempera aspects of speech,GAUSSIAN MIXTURE MODELS (GMMS): Intihned ounMFCC MODEL*SYAVETES (NEURAL NEI WORKS) beorsic Foemy §; (Sse CaeDEEP me APPROACHES (NEURAL NETWORKS) Jorn hong Ns ed Eat “Sle jas] fem a AacktSoul eo! (iztfean)MFCC MODELey eeVECTOR QUANTIZATION (VQ): if i+ ane ee i 3 aaHIDDEN MARKOV MODELS (HMMS): HMMs in ASR Originally used in speech recognition (Rabiner, 1986) Proposed for DNA modeling (Churchill, 1989) Applied to modeling proteins (Haussler et al, 1992) Multiple sequence alignment Ieee le dise\eeM cull Aull (‘homologs’)tees yea freemesoe Analog Circuits vs Digital CircuitsBasic terms (i fie eFe Sirs) Burt incecuktsdue Gainfice (Pa Sais ris S775 enre Siete date SUT) (ierch= ia].SUIsignal is continuous and time varying. Troubleshooting of analog signals are difficult. ‘An analog signal is usually in the form of sine wave. | asly affected by the noise ‘Analog signals use continous values to represent the data, curacy of the analog signals may be affected by noise. |Analog signals may be affected during data transmission, log signal u fore power, Examples: Temperature, Pressure, Flow measurements, ete, |Components like resistors, Capacitors, Inductors, Diodes are | used in analog circuits. SS idios and Vide ANALOG 0s: DIGITAL & Digital signal have two or more states and in binary form. ‘Troubleshooting of digital signals are easy, |An digital signal is usually in the form of square wave, ‘These are stable and less prone to noise. Digital signals use discrete values to represent the data, ‘Accuracy of the digital signals are immune. from the noise. Digital signals are not affacted during data transmission, Digital signal use less power. Examples: Valve Feedback, Motor Start, Trip, ete. Components like transistors, logic gates, and microcontrollers are used in Digital circuits, |and what an R, Land C does ina mp into the main topics lets unders re denoted by the letter “R”. A resistor is an element that mostly in form of heat. It will have a Voltage drop across it which mains fixed for a fixed value of current flowing through it * Capacitor: Capacitors are denoted by the letter “C”. A capacitor is an element ich stores energy (temporarily) in form of electric field. Capacitor resists chang, i ge. There are many types of capacitors, out of which the ceramic capacitor C tic capacitors are mostly used. They charge in one direction and ‘ opposite direction s are denoted by the letter “L”. A Inductor is also similar to y but is stored in form of magnetic field. Inductors + Resistor: Resistors a d s ener . Inductors are normally a coil wound wire and is rarely used er two components. r esistor, Capacitor and Inductors are put together we can form circuits . RLand RLC circuit which exhibits time and frequency dependent responses 1 le usef circuit can k every is tutorial n. many AC applications as mentioned already. A RC/RL/RLC used asa filter, oscillator and much more it is not possible to pect in this tutorial, so we will learn the basic behaviour of them in¢ RC circuit: ¢ The RC circuit (Resistor Capacitor Circuit) will consist ofa Capacitor and a Resistor connected either in series or parallel to a voltage or current source. These types of circuits are also called as RC filters or RC networks since they are most commonly used in filtering applications. An RC circuit can be used to make some crude filters like low-pass, high-pass and Band-Pass filters. c ae \| I Te RC Circuit —————+ RL circuit: + The RL Circuit (Resistor Inductor Circuit) will consist of an Inductor and a Re: again connected either in series or parallel. A series RL circuit will be driven by voltage source and a parallel RL circuit will be driven by a current source. RL circuii commonly used in as passive filters, a first order RL uit with only one inductor stor ir and one capacitor is shown below + Similarly ina RL cireuit we have to replace the Capacitor with an Inductor. The Light sumed to act as a pure resistive load and the resistance of the bulb is set toa bulb is known value of 100 ohms.ARLC circuit as the name implies will consist of a Resistor, Capacitor and Inductor connected in series or parallel. The circuit forms an Oscillator circuit which is very commonly used in Radio receivers and televisions. It is also very commonly used as damper circuits in analog applications. The resonance property of a first order RLC circuit is discussed below The RLC circuit is also called as series resonance circuit, oscillating circuit or a tuned circuit. RLC Circuit