A deep learning approach for high resolution BRIRs generation for real-time AVR

A deep learning approach for high resolution BRIRs generation for real-time AVR

Sanaguano-Moreno, D.A. Departamento de Informática y Ciencias de la Computación, Escuela Politecnica Nacional, Quito, Pichincha, Ecuador, daniel.sanaguano@epn.edu.ec
Lucio-Naranjo, J.F. Departamento de Informática y Ciencias de la Computación, Escuela Politecnica Nacional, Quito, Pichincha, Ecuador, jose.lucio@epn.edu.ec
Tenenbaum, R.A. Pós-Graduação em Engenharia Civil, Universidade Federal de Santa Maria, Santa Maria, RS, Brasil, ratenenbaum@gmail.com
Regattiere-Sampaio, G.B. Engenharia Acústica, Universidade Federal de Santa Maria, Santa Maria, RS, Brasil, gabriel.sampaio@eac.ufsm.br

Abstract

Generating a high-resolution binaural room impulse responses (BRIRs) for a given position requires a significant amount of computational resources, making it impractical for acoustic virtual reality in real-time environments. The main problem is the thousands of coefficients that high-resolution BRIRS possess, even in low-reverberant environments. In this sense, this work introduces a novel methodology for the generation of high-resolution BRIRS from receiver positional information with a computational cost suitable for real-time applications using a deep learning approach. The methodology consists of 1) generating a dataset with enough high-precision BRIRS distributed over a given scenario; 2) training variational autoencoder to obtain a dimensionally reduced BRIRS from the encoder; 3) training a neural network using the position of the acoustic receiver and the distance to the sound source as inputs and the reduced BRIRS as outputs; 4) apply the decoder of the variational autoencoder to reconstruct BRIRs with full spectral information; 5) the reconstructed BRIR for the receptor position is convolved by segments with an anechoic sound signal to produce the sensation of spatial acoustic effect. In this preliminary work, the methodology has been applied until the learning of reduced BRIRS. Also, segments of RIRs were used instead of the complete BRIRS for sake of technical viability. To test the results, the mean square error between the RIRS simulated by the acoustic software and the RIRs generated by the model were compared. Finally, computational cost to generate the RIR has been analyzed with respect to the total number of operations.

Keywords:  
Real-time Auralization
,
Acoustic Virtual Reality
,
Variational Auto-encoders.

Resumen

Generar una respuestas impulsiva biauricular de sala (RIBSS) de alta definición para una posición determinada requiere una cantidad significativa de recursos computacionales, siendo poco práctico para realidad virtual acústica. El principal problema son los miles de coeficientes que poseen las RIBSs de alta resolución, incluso en entornos de baja reverberación. En este sentido, este trabajo presenta una metodología para la generación de RIBSS de alta resolución a partir de información posicional del receptor acústico a un costo computacional adecuado para entornos en tiempo real utilizando un enfoque de aprendizaje profundo. La metodología consiste en: 1) generar un conjunto de datos con suficientes RIBSS de alta resolución distribuidos en un escenario virtual definido; 2) entrenar el autoencoder variacional para obtener RIBSS reducidas en dimensión a partir del codificador; 3) entrenar una red neuronal probabilística utilizando la posición del receptor acústico y la distancia euclidiana a la fuente de sonido como entradas; y las RIBSs reducidas como salidas; 4) aplicar el decodificador del autoencoder variacional para reconstruir RIBSS con la información espectral completa; 5) la RIBS reconstruida para la posición del receptor se convoluciona por segmentos con una señal sonora anecoica para producir la sensación acústica espacial. En este trabajo preliminar, se implementó la metodología hasta la obtención de las RIBSS reducidas al entrenar el autoencoder variacional. En aras de la viabilidad técnica y tecnológica, se utilizaron segmentos de respuestas impulsivas de sala (RISS) en lugar de RIBSs. Para evaluar el sistema se comparo el error cuadrático medio entre las RISS simuladas por un software acústico y las RISS generadas por el modelo. Finalmente, se realizó un análisis de costo computacional de la generación de RISS.

Palabras clave:  
Aurilización tiempo-real
,
Realidad Virtual Acústica
,
Variational Auto-encoders.

Referencias

[1] Tommasini, Fabián C.; Ramos, Oscar A.; Hüg, Mer cedes X. e Ferreyra, Sebastián P. A Compu tational Model to Implement Binaural Synthesis in a Hard Real-Time Auditory Virtual Environment. Acoust. Aust., 47(1):51–66, 2019. ISSN 18392571. doi:10.1007/s40857-019-00152-7. Disponível em: https://doi.org/10.1007/s40857-019-00152-7.
[2] Vorländer, Michael. Virtual Acoustics. 39(3):307 318, 2014. ISSN 01375075. doi:10.2478/aoa-2014 0036.
[3] Schröder, Dirk; Pelzer, Sönke; Vetterli, Martin e Vor länder, Michael. Through the hourglass: A faithful audiovisual reconstruction of the old montreux ca sino. Acoust. Aust., 43(1):49–57, 2015. ISSN 08146039. doi:10.1007/s40857-015-0009-4. Dis ponível em: http://dx.doi.org/10.1007/s40857-015 0009-4.
[4] Pulkki, Ville e Svensson, U. Peter. Machine learning-based estimation and rendering of scat tering in virtual reality. J. Acoust. Soc. Am., 145(4):2664–2676, 2019. ISSN 0001-4966. doi:10.1121/1.5095875. Disponível em: http: //dx.doi.org/10.1121/1.5095875.
[5] Vorländer, Michael e Summers, Jason E. Auraliza tion: Fundamentals of acoustics, modelling, simu lation, algorithms, and facoustic virtual reality. J. Acoust. Soc. Am., 123(6):4028, 2008.
[6] Tenenbaum, R A; Taminato, F O; Melo, V S G e Torres, J C B. Auralization generated by modeling HRIRs with artificial neural networks and its valida tion using articulation tests. Appl. Acoust., 130:260 269, 2018. doi:10.1016/j.apacoust.2017.09.025.
[7] Bork, Ingolf. A comparison of room simulation software-the 2nd round robin on room acoustical computer simulation. Acta Acustica united with Acustica, 86(6):943–956, 2000.
[8] Bianco, Michael J; Gerstoft, Peter; Traer, James; Ozanich, Emma; Roch, Marie A; Gannot, Sharon e Deledalle, Charles-Alban. Machine learning in acoustics: Theory and applications. J. Acoust. Soc. Am., 146(5):3590–3628, 2019.
[9] Naranjo, José F Lucio e Tenenbaum, Roberto A. Inteligˆencia computacional aplicada na geraçao de respostas impulsivas bi-auriculares e em aurilizacao de salas. Tese de doutorado, Universidade do Estado do Rio de Janeiro, 2014.
[10] Taminato, Filipe Otsuka. Redes neurais artificiais aplicadas à modelagem de respostas impulsivas as sociadas à cabeça humana para gerar aurilização. Tese de doutorado, Universidade do Estado do Rio de Janeiro, 2018.
[11] Tommasini, Fabián C.; Ramos, Oscar A.; Hüg, Mer cedes X. e Bermejo, Fernando. Usage of spectral distortion for objective evaluation of personalized HRTF in the median plane. International Journal of Acoustics and Vibrations, 20(2):81–89, 2015. ISSN 10275851. doi:10.20855/ijav.2015.20.2371.
[12] Tamulionis, Mantas e Serackis, Arturas. Compa rison of Multi-Layer Perceptron and Cascade Feed Forward Neural Network for Head-Related Trans fer Function Interpolation. 2019 Open Conferen ce of Electrical, Electronic and Information Sciences, eStream 2019- Proceedings, págs. 1–4, 2019. doi:10.1109/eStream.2019.8732158.
[13] Lluis, Francesc; Martinez-Nuevo, Pablo; Bo Møller, Martin e Ewan Shepstone, Sven. Sound field recons truction in rooms: Inpainting meets super-resolution. J. Acoust. Soc. Am., 148(2):649–659, 2020.
[14] Karakonstantis, Xenofon e Fernandez Grande, Efren. Sound field reconstruction in rooms with deep gene rative models. Em INTER-NOISE and NOISE-CON Congress and Conference Proceedings, volume 263, págs. 1527–1538. Institute of Noise Control Engi neering, 2021.
[15] Kingma, Diederik P e Welling, Max. An intro duction to variational autoencoders. arXiv preprint arXiv:1906.02691, 2019.
[16] Blundell, Charles; Cornebise, Julien; Kavukcuoglu, KorayeWierstra, Daan. Weightuncertainty in neural network. Em International conference on machine learning, págs. 1613–1622. PMLR, 2015.
[17] Blei, David M; Kucukelbir, Alp e McAuliffe, Jon D. Variational inference: A review for statisticians. J AmStat Assoc, 112(518):859–877, 2017.
[18] Sinitambirivoutin, Emrick. An introduction to Va riational Auto Encoders (VAEs). https://toward sdatascience.com/an-introduction-to-variational auto-encoders-vaes-803ddfb623df, 2020. [Online; accessed 8-Abril-2022].
[19] Tenenbaum, Roberto A.; Taminato, Filipe O. e Melo, Viviane S.G. Fast auralization using radial basis functions type of artificial neural network techniques. Appl. Acoust., 157:106993, 2020. ISSN 0003682X. doi:10.1016/j.apacoust.2019.07.041. Disponível em: https://doi.org/10.1016/j.apacoust.2019.07.041.
[20] Lindau, Alexander. The perception of system latency in dynamic binaural synthesis. Proc. of 35th DAGA, págs. 1063–1066, 2009.
[21] Mackensen, Philip. Auditive localization. head movements, an additional cue in localization. PhD dissertation, Tech. Univ., 2004.
[22] Yairi, Satoshi; Iwaya, Yukio e Suzuki, Yôiti. In vestigation of system latency detection threshold of virtual auditory display. Georgia Institute of Techno logy, 2006.
Publicado em:
01/09/2022