text stringlengths 1 6.06k | source stringclasses 375
values | page int64 0 1.49k | book stringclasses 375
values | chunk_index int64 0 0 |
|---|---|---|---|---|
varient de 4.5% (DL) à 7.0% (IL). Bien que ces résultats soient plus mauvais que ceux obtenus sur des tâches apparemment plus complexes (par exemple, 3% sur RM), ils restent relativement acceptables étant donné le haut degré de confusion des différents mots de vocabulaire (accentué par la bande passante téléphonique li... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 137 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 129 où les variabilités dues aux facteurs signalés dans la Section (1.1) sont plus importantes. 18.2 Applications La technologie de reconnaissance n’est donc pas encore parfaite et souffre encore des nombreux facteurs de variabilité relatifs aux conditions réelles d’utilisatio... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 138 | 2 intro_ASR_f | 0 |
dictée vocaux. Particulièrement ambitieuse, cette dernière application n’est cependant pas entièrement hors de portée lorsque l’on considère que, non seulement des systèmes de laboratoire, mais également des produits commerciaux, sont actuelle- ment disponibles (et fonctionnent en temps réel sur PC) et permettent de re... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 138 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 130 7. Applications éducatives: par exemple, en facilitant l’apprentissage de langues étrangères (étant donné qu’un système de reconnaissance ne fonctionnera bien que lorsque l’accent correct est utilisé) ou l’interaction vocale (plus attrayante et plus facile) avec des progra... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 139 | 2 intro_ASR_f | 0 |
thmes efficaces (en termes de mémoire et CPU) dont la mise en oeuvre a été fortement optimisée. Les algorithmes de recherche actuels, utilisant notamment des méthodes d’éla- guage sophistiquées, permettent en effet de développer des systèmes de reconnaissance de parole continue grands vocabulaires (par exemple, 60000 m... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 139 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 131 2. La qualité des modèles HMM sous-jacents, incluant notamment les hypothèses rela- tives à la topologie des modèles et les hypothèses concernant les densités de proba- bilité. Finalement, pendant la phase de reconnaissance, le système calculera simplement l’explica- tion ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 140 | 2 intro_ASR_f | 0 |
quelques centaines de mots, mots isolés ou parole continue, indépendants du locuteur). Aujourd’hui, il est même possible de démontrer des systèmes de reconnaissance de parole continue indépendants du locuteur et pouvant reconnaître un vocabulaire de 60000 mots, 75 et ceci avec un taux de reconnaissance avoisinant les 9... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 140 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 132 | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 141 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 133 Références [1] Abeles, M., Corticonics: Neural Circuits of the Cerebral Cortex, Cambridge University Press, 1991. [2] Atal, B.S., Automatic speaker recognition based on pitch contours, Journal of Acous- tical Society of America, vol. 52, pp. 1687-1697, 1972. [3] Bahl, L.R.... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 142 | 2 intro_ASR_f | 0 |
De Mori, R., Flammia, G., & Kompe, R., Global optimization of a neural network-Hidden Markov Model hybrid, IEEE Trans, on Neural Networks, vol. 3, no. 2, pp. 252-259, 1992. [10] Blahut, R., Principles and Practice of Information Theory, Addison-Wisley, 1987. [11] Boite, R. & Kunt, M., Traitement de la Parole, Presses P... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 142 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 134 [17] Bridle, J.S., Brown, M.D., & Chamberlain, R.M., An Algorithm for Connected Word Recognition, Proc. IEEE Intl. Conf. on Acoustic, Speech, and Signal Processing (Pa- ris), pp.899-902, 1982. [18] Bridle, J.S., Alpha-Nets: A Recurrent “Neural” Network Architecture with a ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 143 | 2 intro_ASR_f | 0 |
word recognition in continuously spoken sentences, IEEE Trans. Acoustics, Speech, and Signal Processing, vol. 28, no. 4, pp. 357-366, 1980. [26] Dempster, A.P., Laird, N.M., & Rubin, D.B., Maximum Likelihood From Incomplete Data Via the EM Algorithm, Journal of the Royal Statistical Society, vol. 39, pp. 1-38, 1977. [2... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 143 | 2 intro_ASR_f | 0 |
. [34] Furui, S., Towards robust speech recognition under adverse conditions, Proceedings of the ESCA Workshop on Speech Processing and Adverse Conditions (Cannes- Mandelieu, France), pp. 31-41, 1993. | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 143 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 135 [35] Furui, S., An overview of speaker recognition technology, in Automatic Speech and Speaker Recognition, C.-H. Lee, F.K. Soong, K.K. Paliwal (eds.), pp. 31-56, Kluwer Academic Publishers, 1996. [36] Genoud, D., Moreira, M., and Mayoraz, E., Text dependent speaker verifi... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 144 | 2 intro_ASR_f | 0 |
, vol. 2, pp. 83-86, 1993. [42] Higgins, A.L., Bahler, L., & Porter, J., Speaker verification using randomized phrase prompting, Digital Signal Processing, vol. 1, pp. 89-106, 1991. [43] Huang, X., Lee, K., & Waibel, A., Connectionist speaker normalization and its appli- cation to speech recognition, Proc. of IEEE Work... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 144 | 2 intro_ASR_f | 0 |
de neurones récursifs pour mémoires associatives, Presses Polytechniques et Universitaires Romandes, 1990. [51] Katagiri, S., Lee, C.-H., & Juang, B.H., New discriminative training algorithms based on the generalized probabilistic descent method, IEEE Proc. Workshop on Neural Networks for Signal Processing, B.H. Juang,... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 144 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 136 [52] Kehagias, A., Optimal Control for Training: The Missing Link Between Hidden Mar- kov Models and Connectionist Networks, Division of Applied Mathematics Technical Report, Brown University, Providence (RI), 1989. [53] Kleinrock, L., Queuing Systems, 2 vols., Wiley, New ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 145 | 2 intro_ASR_f | 0 |
, P.C., Maximum likelihood linear regression for speaker adaptation of continuous density HMMs, Computer, Speech and Language, vol. 9, pp. 171-186, 1995. [61] Levin, E., Speech recognition using hidden control neural network architecture, in Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Albuquerqu... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 145 | 2 intro_ASR_f | 0 |
of the National Inst. Sci. (India), vol. 12, pp. 49-55, 1936. [68] Matsui, T. & Furui, S., Text-independent speaker recognition using vocal tract and pitch information, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Albuquerque, NM), pp. 137-140, 1990. | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 145 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 137 [69] Matsui, T. & Furui, S., Concatenated phoneme models for text-variable speaker re- cognition, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Min- neapolis, MI), pp. II-391-394, 1993. [70] Montacie, C., Deleglise, P., Bimbot, F., & Caraty, M.-J., Ci... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 146 | 2 intro_ASR_f | 0 |
y, H., The use of a one-stage dynamic programming algorithm for connected word recognition, IEEE Trans. on Acoustics, Speech, and Signal Processing, vol. 32, no. 2, pp. 263-271, April 1984. [77] Ney, H. & Aubert, X., Dynamic programming search strategies: From digit strings to large vocabulary word graphs, in Automatic... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 146 | 2 intro_ASR_f | 0 |
perceptron, Speech Communication, vol. 7, pp. 403-409, 1988. [84] Poritz, A.B., Linear predictive hidden Markov models and the speech signal, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Paris, France), pp. 1291- 1294, 1982. | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 146 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 138 [85] Rabiner, L.R., A tutorial on hidden Markov models and selected applications in speech recognition, Proceedings of the IEEE, vol. 77, no. 2, pp. 257-285, 1989. [86] Rabiner, L. & Juang, B.-H., Fundamentals of Speech Recognition, Englewood Cliffs, N.J.: Prentice Hall, 1... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 147 | 2 intro_ASR_f | 0 |
1990. [92] Rosenberg, A.E., Lee, C.-H., & Gokcen, S., Connected word talker verification using whole word hidden Markov models, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Toronto, Canada), pp. 381-384, 1991. [93] Rosenberg, A.E., DeLong, J., Lee, C.-H., Juang, B.H., and Soong, F.K., The use of ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 147 | 2 intro_ASR_f | 0 |
Acoustics, Speech, and Signal Processing (Albuquerque, NM), pp. 281-284, 1990. [99] Schwartz, R., Nguyen, L., & Makhoul, J., Multiple-pass search strategies, Automatic Speech and Speaker Recognition, C.-H. Lee, F.K. Soong, K.K. Paliwal (eds.), pp. 429- 456, Kluwer Academic Publishers, 1996. [100] Seneff, S., A joint sy... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 147 | 2 intro_ASR_f | 0 |
Reconnaissance de la parole et du locuteur 139 [101] Soong, F.K., Rosenberg, A.E., Rabiner, L.R., & Juang, B.-H., A vector quantization approach to speaker recognition, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Tampa, FL), pp. 387-390, 1985. [102] Sorenson, H., A cepstral noise reduction multi... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 148 | 2 intro_ASR_f | 0 |
édition, Addison- Wesley, Reading, Mass., 1980. [108] Varga, A.P., & Moore, R.K., Hidden Markov model decomposition of speech and noise, Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing (Albu- querque, NM), pp. 845-848, 1990. [109] Vintsyuk, T., Speech discrimination by dynamic programming, Kibernetik... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/2 intro_ASR_f.pdf | 148 | 2 intro_ASR_f | 0 |
Advanced Probability and Applications EPFL - Fall Semester 2024-2025 Solutions to Homework 12 Exercise 1. a) We must check that E(ψ(Y ) g(Y )) = E(X g(Y )) for any continuous and bounded function g. The computation gives indeed: E(ψ(Y ) g(Y )) = X y∈C ψ(y) g(y) P({Y = y}) = X x,y∈C x g(y) P({X = x, Y = y}) = E(X g(Y ))... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/sol12_1.pdf | 0 | sol12_1 | 0 |
E(Z|N)) = E(pN) = pλ. b) We have that P({N = k}|{Z = z}) = P({N = k, Z = z}) {Z = z}) = <unk>k z pzqk−z(λk/k!)e−λ P m≥z <unk>m z pzqm−z(λm/m!)e−λ = (qλ)k−z (k −z)!e−qλ From the previous Exercise, ψ(z) = X k≥0 kP({N = k}|{Z = z}) = n X k≥z k (qλ)k−z (k −z)!e−qλ = z + qλ and so E(N|Z) = Z + qλ. c) Intuitively, we would l... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/sol12_1.pdf | 0 | sol12_1 | 0 |
Exercise 3. a) Let us compute theoretically the first three MSE’s: E(( b X1 −X)2) = E X + Z a −X 2! = E(Z2) a2 = 1 a2 E(( b X2 −X)2) = E a2X + aZ a2 + 1 −X 2! = E − 1 a2 + 1 X + a a2 + 1 Z 2! = 1 (a2 + 1)2 + a2 (a2 + 1)2 = 1 a2 + 1 E(( b X3 −X)2) = E((sign(a2X + aZ) −X)2) = 4Q(−|a|) where Q(x) = R x −∞pZ(z)dz is the cd... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/sol12_1.pdf | 1 | sol12_1 | 0 |
−ay eay + e−ay = tanh(ay) which confirms that E(X|Y ) = tanh(aY ) = b X4. NB: The first expression for the function ψ(y) above can be obtained either by reasoning intuitively (and forgetting that we are dealing here with a mix of discrete (X) and continuous (Y ) random variables), or by proving formally that the random... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/sol12_1.pdf | 1 | sol12_1 | 0 |
c) Using the following series of equalities: E((E(X|Y ) −X)2) = E(X2) + E(E(X|Y )2) −2E(XE(X|Y )) = E(X2) + E(E(X|Y )2) −2E(E(XE(X|Y )|Y )) = E(X2) + E(E(X|Y )2) −2E(E(X|Y )2) = E(X2) −E(E(X|Y )2) we see that E(( b X4 −X)2) = E(X2) −E( b X2 4) = 1 −E(tanh(aY )2) = 1 −E(tanh(a2 + aZ)2) (noticing for the last equality th... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/sol12_1.pdf | 2 | sol12_1 | 0 |
Spatial Hearing Introduction Christof Faller November 21, 2017 Audiovisual Communications Laboratory, EPFL, Switzerland | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 0 | 7 spatial_hearing | 0 |
Contents –Introduction –Spatial Hearing in Free Field –Spatial Hearing in Rooms –Conclusions More details and references: Lecture Notes: "Signal Processing for Audio and Acoustics" C. Faller | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 1 | 7 spatial_hearing | 0 |
Introduction Perception of an auditory spatial image | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 2 | 7 spatial_hearing | 0 |
Contents –Introduction –Spatial Hearing in Free Field –Spatial Hearing in Rooms –Conclusions | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 3 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field One sound source | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 4 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field One sound source –"Sine law" for spatial hearing (Hornbostel & Wertheimer 1920) –More accurate path-length difference formulas (Blauert 1997) ∆d = κ sin ⇥with κ = 21 cm | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 5 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field One sound source –Interaural Time Difference (ITD) ITD = arg max d E{e1(t)e2(t + d)} E{x(t)} = lim T⇥⇤ 1 2T ⇤T −T x(t)dt with | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 6 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field One sound source –Interaural Level Difference (ILD) ILD = 10 log10 !E{e2 2(t)} E{e2 1(t)} ⇥ Head shadowing | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 7 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field One sound source –Relation between source direction and ITD/ILD | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 8 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Ear entrance signal properties and lateralization –Experimental setup | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 9 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Ear entrance signal properties and lateralization –ITD/ILD | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 10 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Ear entrance signal properties and lateralization –Duplex Theory | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 11 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Ear entrance signal properties and lateralization –Interaural Coherence (IC) IC = max d |E{e1(t)e2(t + d)}| ! E{e2 1(t)}E{e2 2(t + d)} | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 12 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Ear entrance signal properties and lateralization –Interaural Coherence (IC) DEMO coherence | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 13 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Two sound sources: summing localization –Inter-Channel Time Difference (ICTD) –Inter-Channel Level Difference (ICLD) | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 14 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Two sound sources: summing localization –Inter-Channel Coherence (ICC) DEMO coherence | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 15 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Superposition of one auditory object ear entrance signals –Concurrent independent sources DEMO superposition | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 16 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Superposition of one auditory object ear entrance signals –Concurrent independent virtual sources | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 17 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Superposition of one auditory object ear entrance signals –Concurrent independent lateralized sources | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 18 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Cone of Confusion | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 19 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Monaural Spatial Cues – "Blauert Bands" | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 20 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Monaural Spatial Cues – "Blauert Bands" Figure source: Wikipedia.de Probability of direction detection in % Frequency in kHz DEMO Blauert Bands back front top front back Direction Determining Bands | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 21 | 7 spatial_hearing | 0 |
Spatial Hearing in Free Field Monaural Spatial Cues – Head Movements | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 22 | 7 spatial_hearing | 0 |
Contents –Introduction –Spatial Hearing in Free Field –Spatial Hearing in Rooms –Conclusions | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 23 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Image method | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 24 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Perceptually motivated parts | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 25 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Reverberation time RT60 (Sabine 1922) 60dB RT60 TIME SOUND LEVEL DEMO reverberation time | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 26 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Reverberation time computed from RIR (Schroeder 1965) σ2 y = σ2 s ! ∞ 0 h2(t)dt σ2 y(⇥) = σ2 s ! ∞ τ h2(t)dt Microphone signal: Steady state sound level: Sound level after source stopped: y(t) = h(t) ? s(t) = Z 1 0 h(<unk>)s(t −<unk>)d<unk> | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 27 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Reverberation time computed from RIR (Schroeder 1965) Relative level after sound stopped: a(⇥) = 10 log10 σ2 y(⇥) σ2y(0) = 10 log10 ! ∞ τ h2(t)dt ! ∞ 0 h2(t)dt = -60dB RT60 | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 28 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Measuring Y (⇤) = H(⇤)S(⇤) H(⇤) = Y (⇤) S(⇤) y(t) = h(t) <unk>s(t) | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 29 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Room impulse response (RIR) –Measuring: Sweeps s(t) = A sin( ⇤t 0 ⇥(τ)dτ) Linear: Logarithmic: ⇥(τ) = ⇥0 + τ T (⇥1 −⇥0) ⇥(τ) = elnω0+ τ T (lnω1−lnω0) = ⇥0 "⇥1 ⇥0 ⇥τ T | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 30 | 7 spatial_hearing | 0 |
0 50 100 150 200 −1 −0.5 0 0.5 1 TIME [ms] Amplitude (a) 0 50 100 150 200 −1 −0.5 0 0.5 1 TIME [ms] Amplitude (b) Spatial Hearing in Rooms Room impulse response (RIR) –Measuring: Sweeps Linear Logarithmic DEMO sweeps | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 31 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Source localization in the presence of reflections –Precedence effect (Zurek 1987, etc.) DEMO precedence effect | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 32 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Source localization in the presence of reflections –Three phases of the precedence effect (Blauert 1997) | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 33 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Source localization in the presence of reflections –Echo threshold (Damaske 1971, Meyer & Schodder 1979) 0 10 20 30 40 50 60 70 80 90 100 −30 −20 −10 0 10 DELAY [ms] LEVEL DIFFERENCE [dB] (a) 0 10 20 30 40 50 60 −25 −20 −15 −10 −5 0 5 10 15 DELAY [ms] LEVEL DIFFERENCE [dB] (b) 10ms 30ms 100ms S... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 34 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Coloration (effect of early reflections) Example: 0 0.5 1 h1 −10 −5 0 5 10 |H1| [dB] 0 5 10 15 20 TIME [ms] 0 0.5 1 1.5 2 10 FREQUENCY [kHz] DEMO coloration | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 35 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Coloration (effect of early reflections) Example: 0 5 10 15 20 0 0.5 1 TIME [ms] h2 0 0.5 1 1.5 2 −10 −5 0 5 10 |H2| [dB] FREQUENCY [kHz] DEMO coloration | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 36 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Distance of auditory object Distance perception is related to: –Absolute sound level –Direct to reverberant ratio –Knowledge of source level (e.g. speech) –Near field cues (only for sources very near to listener) | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 37 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Width of auditory object (ASW) LF80 5 = ! 80ms 5ms h2(t) cos2 α(t)dt ! 80ms 0ms h2(t)dt Early lateral fraction: | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 38 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Width of auditory object (ASW) Experiment: DEMO asw | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 39 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Width of auditory object (ASW) Late lateral fraction: LF∞ 80 = ! ∞ 80ms h2(t) cos2 α(t)dt ! ∞ 0ms h2(t)dt | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 40 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Spatial Impression –Listener Envelopment (LE) Experiment: DEMO le | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 41 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Binaural Room Impulse Responses (BRIRs) | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 42 | 7 spatial_hearing | 0 |
Spatial Hearing in Rooms Physical measures for width and evelopment related to BRIRs –Interaural cross-correlation coefficient (IACC) –Bradley (1994), Okano et al. (1998) IACC = max τ Φ(τ) Φ(τ) = ! T2 T1 hl(t)hr(t + τ)dt ⇥! T2 T1 h2 l (t)dt ! T2 T1 h2r(t)dt with Measure BRIR segment IACC(E) 0 −80 ms IACC(L) 80 −1000 ms... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 43 | 7 spatial_hearing | 0 |
Conclusions Summarized basic spatial hearing phenomena. | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/7 spatial_hearing.pdf | 44 | 7 spatial_hearing | 0 |
CS-472: Design Technologies for Integrated Systems Exercise Problem Set 1 Solution Date: 30/09/2021 Problem 1 Given the graph G(V, E) below: v0 v1 v2 v3 v4 v5 v6 cf: Textbook pp.40-42. (a) Color the graph with the smallest number of colors. Ans: v0 v1 v2 v3 v4 v5 v6 (b) Show a minimum clique cover. Ans: (For example,) ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/ex1_sol.pdf | 0 | ex1_sol | 0 |
4. The clique cover number κ(G) = 3 (i.e. size of the minimum clique cover, as in (b)). ω(G) = χ(G) and α(G) = κ(G), so G is a perfect graph. (e) Draw the complement graph. (f) Color the complement graph with the smallest number of colors. Ans: (d)+(e). Note that this coloring is also the solution of (c). v0 v1 v2 v3 v... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/ex1_sol.pdf | 1 | ex1_sol | 0 |
Problem 2 Given the directed acyclic graph G(V, E, W) below: v0 v1 v2 v3 v4 v5 v6 4 2 3 1 2 2 3 2 1 Find the shortest path from the source v0 to the sink v6 by applying the following algo- rithms: cf: Textbook pp.55. (a) Dijkstra algorithm. vq = v0 v2 v1 v3 v5 v4 v6 s0 0 0 0 0 0 0 0 s1 4 4 4 4 4 4 4 s2 2 2 2 2 2 2 2 s3... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/ex1_sol.pdf | 2 | ex1_sol | 0 |
Exercise Set III, Advanced Algorithms 2022 These exercises are for your own benefit. Feel free to collaborate and share your answers with other students. This exercise set contains many problems. So solve as many problems as you can and ask for help if you get stuck for too long. Problems marked * are more difficult bu... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 0 | Exercises3-sol_1 | 0 |
x∗ v −ε if v ∈V + x∗ v + ε if v ∈V − x∗ v otherwise Note that x∗= 1 2y+ + 1 2y−. It remains to verify that y+ and y−are feasible solutions. 1. By selecting ε small enough, the boundary constraints (0 ≤y+ v ≤1, 0 ≤y− v ≤1) are satisfied. Page 1 (of 4) Advanced Algorithms • Spring 2022 Michael Kapralov and Ola Svensson | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 0 | Exercises3-sol_1 | 0 |
2. Consider the constraints for the edges e = {u, v} ∈E. If x∗ u + x∗ v > 1, the constraint remains satisfied by picking ε > 0 small enough. If x∗ u +x∗ v = 1, then consider the following cases: • u, v /∈V + ∪V −. In this case, y+ u + y+ v = x∗ u + x∗ v = 1. • u ∈V +; then v ∈V −. In this case, y+ u + y+ v = x∗ u + ε +... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 1 | Exercises3-sol_1 | 0 |
2 and y1 +4y2 ≥13 for x3. It follows that we can formulate the problem of finding an upper bound as the following linear program (the dual): Minimize 24y1 + 60y2 Subject to y1 + y2 ≥6 3y1 + 2y2 ≥14 y1 + 4y2 ≥13 y1, y2 ≥0 3 Consider the min-cost perfect matching problem on a bipartite graph G = (A ∪B, E) with costs c : ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 1 | Exercises3-sol_1 | 0 |
Solution: Let vb = 0 for all b ∈B and ua = min{a,b}∈E c({a, b}) be a dual solution. By definition it is feasible. Now define the vector (u∗, v∗) by u∗ a = ( 1 if a ∈S 0 otherwise and v∗ b = ( −1 if b ∈N(S) 0 otherwise Note that (u, v) + α · (u∗, v∗) is a feasible solution for any scalar α ≥0. Such a solution has dual v... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 2 | Exercises3-sol_1 | 0 |
should be n. To this end, let A0 = {a0} and B0 = N(a0). Note that all vertices of B0 are covered by the matching M (if b0 ∈B0 is not covered, the edge a0b0 can be added to the matching which contradicts the fact that M is a maximum matching). If B0 = ∅, S = A0 is a set such that |N(S)| < |S|. Else, B0 is matched with |... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 2 | Exercises3-sol_1 | 0 |
path p is used and 0 otherwise2. Let P be the set of all such paths 1Some parts of this proof are taken from this link. 2I know that the number of variables may be exponential, but let us not worry about that. Page 3 (of 4) Advanced Algorithms • Spring 2022 Michael Kapralov and Ola Svensson | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 2 | Exercises3-sol_1 | 0 |
from s to t. The linear programming relaxation of this problem now becomes Maximize X p∈P xp subject to X p∈P:e∈p xp ≤1, ∀e ∈E, xp ≥0, ∀p ∈P. What is the dual of this linear program? What famous combinatorial problem do binary solutions to the dual solve? Solution: The dual is the following: minimize X e∈E ye subject t... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises3-sol_1.pdf | 3 | Exercises3-sol_1 | 0 |
Exercise Set X, Advanced Algorithms 2021 These exercises are for your own benefit. Feel free to collaborate and share your answers with other students. Solve as many problems as you can and ask for help if you get stuck for too long. Problems marked * are more difficult but also more fun :). These problems are taken fr... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 0 | Exercises10-sol_1 | 0 |
1 Therefore: • if |pi −qi| > w for some i, then Pr[h(p) = h(q)] = 0, • otherwise Pr[h(p) = h(q)] = d Y i=1 1 −|pi −qi| w ≈ d Y i=1 e−|pi−qi| w = e− Pd i=1 |pi−qi| w = e−∥p−q∥1 w . 1 To see this, assume wlog that pi < qi < pi + w; there will be exactly one bucket-beginning in the interval (pi, pi + w], the position of t... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 0 | Exercises10-sol_1 | 0 |
2 Consider two LSH hash families H1 and H2 designed for a distance function dist : Rd × Rd →R. For r = 0.1 and c = 2, H1 satisfies dist(p, q) <unk>r =⇒Ph∼H1 [h(p) = h(q)] <unk>1/2 dist(p, q) <unk>c · r =⇒Ph∼H1 [h(p) = h(q)] <unk>1/8 and H2 satisfies dist(p, q) <unk>r =⇒Ph∼H2 [h(p) = h(q)] <unk>1/8 dist(p, q) <unk>c · r... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 1 | Exercises10-sol_1 | 0 |
and l∈O(nρ ln n). This yields a space complexity of order O n4/3 ln(n)2 To derive the query time, let us assume that computing dist(p, q) for some p, q ∈Rd takes time O(d) (for instance, this is the complexity of computing the standard euclidean distance). At query time for input q ∈Rd, we need to iterate over the lhas... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 1 | Exercises10-sol_1 | 0 |
Further assume that you have a (r, c · r, p1, p2)-LSH hash family H for the considered distance function with parameters r = 1, c = 2, p1 = 1/2 and p2 = 1/8. That is, dist(p, q) <unk>1 =⇒P [h(p) = h(q)] <unk>1/2 dist(p, q) <unk>2 =⇒P [h(p) = h(q)] <unk>1/8 where the probabilities are over h ∼H. Exploit the sparsity con... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 2 | Exercises10-sol_1 | 0 |
and the space complexity of our approach. It is easy to see that the space complexity and the preprocessing time is exactly same as the one described in the lecture note. For any query point q we need to spend O(l· k) time to compute fi(q) for all 1 ≤i ≤l(we assume the computation of the hash functions take constant ti... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/Exercises10-sol_1.pdf | 2 | Exercises10-sol_1 | 0 |
1 Applied Data Analysis (CS401) Robert West Lecture 9 Unsupervised learning 16 Nov 2022 | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 0 | 09 - Unsupervised learning | 0 |
Announcements ●Next week’s lecture (Wed 23 Nov) to exceptionally take place in SwissTech Convention Center (auditorium C) ●Project milestone P2 due on Fri 18 Nov 23:59 ○Reminder: we won’t answer questions asked in the final 24 hours before the deadline ●Homework H2 to be released on Fri 18 Nov ○Due two weeks later, on ... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 1 | 09 - Unsupervised learning | 0 |
Give us feedback on this lecture here: https://go.epfl.ch/ada2022-lec9-feedback ●What did you (not) like about this lecture? ●What was (not) well explained? ●On what would you like more (fewer) details? ●... 3 | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 2 | 09 - Unsupervised learning | 0 |
Machine learning • Supervised: We are given input/output pairs (X, y) (a.k.a. “samples”) that are related via a function y = f(X). We would like to “learn” f, and evaluate it on new data. Types: • Discrete y (class labels): “classification” • Continuous y: “regression” (e.g., linear regression) • Unsupervised: Given on... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 3 | 09 - Unsupervised learning | 0 |
◼Given a set of points, with a notion of distance between points, group the points into some number of clusters, such that ▪members of a cluster are close (i.e., similar) to each other ▪members of different clusters are far apart from each other ◼Usually: ▪Points are in a high-dimensional space ▪Similarity is defined v... | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 4 | 09 - Unsupervised learning | 0 |
Characteristics of clustering methods Quantitative: scalability (many samples), dimensionality (many features) Qualitative: types of features (numerical, categorical, etc.), type of shapes (polyhedra, hyperplanes, manifolds, etc.) 6 | /home/ricoiban/GEMMA/mnlp_chatsplaining/RAG/09 - Unsupervised learning.pdf | 5 | 09 - Unsupervised learning | 0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.