Expressive Sampling Synthesis - Learning Extended Source–Filter Models from Instrument Sound Databases for Expressive Sample Manipulations - TEL - Thèses en ligne Accéder directement au contenu
Thèse Année : 2015

Expressive Sampling Synthesis - Learning Extended Source–Filter Models from Instrument Sound Databases for Expressive Sample Manipulations

Synthèse et transformation des sons basés sur les modèles de type source-filtre étendu pour les instruments de musique

Henrik Hahn

Résumé

This thesis addresses imitative digital sound synthesis of acoustically viable instruments with support of expressive, high-level control parameters. A general model is provided for quasi-harmonic instruments that reacts coherently with its acoustical equivalent when control parameters are varied. The approach builds upon recording-based methods and uses signal transformation techniques to manipulate instrument sound signals in a manner that resembles the behavior of their acoustical equivalents using the fundamental control parameters intensity and pitch. The method preserves the inherent quality of discretized recordings of a sound of acoustic instruments and introduces a transformation method that retains the coherency with its timbral variations when control parameters are modified. It is thus meant to introduce parametric control for sampling sound synthesis. The objective of this thesis is to introduce a new general model representing the timbre variations of quasi-harmonic music instruments regarding a parameter space determined by the control parameters pitch as well as global and instantaneous intensity. The model independently represents the deterministic and non-deterministic components of an instrument's signal and an extended source-filter model will be introduced for the former to represent the excitation and resonance characteristics of a music instrument by individual parametric filter functions. The latter component will be represented using a classic source-filter approach using filters with similar parameterization. All filter functions are represented using tensor-product B-splines to support for multivariate control variables. An algorithm will be presented for the estimation of the model's parameters that allows for the joint estimation of the filter functions of either component in a multivariate surface-fitting approach using a data-driven optimization strategy. This procedure also includes smoothness constraints and solutions for missing or sparse data and requires suitable data sets of single note recordings of a particular musical instrument. Another original contribution of the present thesis is an algorithm for the calibration of a note's intensity by means of an analysis of crescendo and decrescendo signals using the presented instrument model. The method enables the adjustment of the note intensity of an instrument sound coherent with the relative differences between varied values of its note intensity. A subjective evaluation procedure is presented to assess the quality of the transformations obtained using a calibrated instrument model and independently varied control parameters pitch and note intensity. Several extends of sound signal manipulations will be presented therein. For the support of inharmonic sounds as present in signals produced by the piano, a new algorithm for the joint estimation of a signal's fundamental frequency and inharmonicity coefficient is presented to extend the range of possible instruments to be manageable by the system. The synthesis system will be evaluated in various ways for sound signals of a trumpet, a clarinet, a violin and a piano.
Dans cette thèse un système de synthèse sonore imitative sera présenté, applicable à la plupart des instruments de quasi-harmoniques. Le système se base sur les enregistrements d’une note unique qui représentent une version quantifiée de l'espace de timbre possible d'un instrument par rapport à sa hauteur et son intensité. Une méthode de transformation permet alors de générer des signaux sonores de valeurs continues des paramètres de contrôle d'expression qui sont perceptuellement cohérent avec ses équivalents acoustiques. Un modèle paramétrique de l'instrument se présente donc basé sur un modèle de filtre de source étendu avec des manipulations distinctes sur les harmoniques d’un signal et ses composantes résiduelles. Une procédure d'évaluation subjective sera présentée afin d’évaluer une variété de résultats de transformation par une comparaison directe avec des enregistrements non modifiés, afin de comparer la perception entre les résultats synthétiques et leur équivalents acoustiques.
Fichier principal
Vignette du fichier
hahn_phd_thesis.pdf (20.72 Mo) Télécharger le fichier
Loading...

Dates et versions

tel-01263656 , version 1 (17-02-2016)

Identifiants

  • HAL Id : tel-01263656 , version 1

Citer

Henrik Hahn. Expressive Sampling Synthesis - Learning Extended Source–Filter Models from Instrument Sound Databases for Expressive Sample Manipulations. Signal and Image Processing. UPMC Université Paris VI, 2015. English. ⟨NNT : ⟩. ⟨tel-01263656⟩
350 Consultations
326 Téléchargements

Partager

Gmail Facebook X LinkedIn More