Gustavo Reis

Genetic Algorithm Approach to Polyphonic Music Transcription

Automatic music transcription (extracting musical notes from a polyphonic audio stream) is a very complex task that continues waiting for solutions, due to the harmonic complexity of musical sounds. Traditional approaches try to extract the information directly from the audio stream, but by taking into account that a polyphonic audio stream is no more than a combination of several notes, music transcription can be considered as a search problem where the goal is to find the sequence of the notes that best models our audio signal. By taking advantage of the genetic algorithms to explore a large search space we present a new approach to the music transcription problem. The results obtained show the feasibility of the approach.

Automatic Transcription of Polyphonic Piano Music Using Genetic Algorithms, Adaptive Spectral Envelope Modeling, and Dynamic Noise Level Estimation

This paper presents a new method for multiple fundamental frequency (F0) estimation on piano recordings. We propose a framework based on a genetic algorithm in order to analyze the overlapping overtones and search for the most likely F0 combination. The search process is aided by adaptive spectral envelope modeling and dynamic noise level estimation: while the noise is dynamically estimated, the spectral envelope of previously recorded piano samples (internal database) is adapted in order to best match the piano played on the input signals and aid the search process for the most likely combination of F0s. For comparison, several state-of-the-art algorithms were run across various musical pieces played by different pianos and then compared using three different metrics. The proposed algorithm ranked first place on Hybrid Decay/Sustain Score metric, which has better correlation with the human hearing perception and ranked second place on both onset-only and onset–offset metrics. A previous genetic algorithm approach is also included in the comparison to show how the proposed system brings significant improvements on both quality of the results and computing time.

Evolving predator and prey behaviours with co-evolution using genetic programming and decision trees

Developing artificial behaviours to control artificial creatures or vehicles is a task that can be solved by means of Evolutionary Algorithms. The Predator and Prey is a problem where it is possible to evolve behaviours for both predator and prey, using artificial co-evolution: the predator must capture the prey and the prey must evade the predator. Both predator and prey have also different characteristics, the predator is faster and more agile and the prey is slower. This paper presents an alternative, using Genetic Programming with Decision Trees for evolving both Predator and Prey behaviours. The results obtained shows the feasibility of the approach.

Electronic synthesis using genetic algorithms for automatic music transcription

This paper presents a novel approach to the problem of automatic music transcription using electronic synthesis with genetic algorithms. Although the problem is well known and different techniques have been applied before, evolutionary algorithms have never been considered when addressing this problem. By means of a series of steps, we show that a polyphonic MIDI file -containing instrument’s partiturescan be automatically generated from an audio recording, by extracting and separating simultaneous notes. The results obtained shows the feasibility of the approach.

A Genetic Algorithm Approach with Harmonic Structure Evolution for Polyphonic Music Transcription

This paper presents a genetic algorithm approach with harmonic structure evolution for polyphonic music transcription. Automatic music transcription is a very complex problem that continues waiting for solutions due to the harmonic complexity of musical sounds. More traditional approaches try to extract the information directly from the audio stream, but by taking into account that a polyphonic audio stream is no more than a combination of several musical notes, music transcription can be addressed as a search space problem where the goal is to find the sequence of notes that best models our audio signal. By taking advantage of the genetic algorithms to explore large search spaces we present a new approach to the music transcription problem. In order to reduce the harmonic overfitting several techniques were used including the encoding of the harmonic structure of the internal synthesizer inside the individuals genotype as a way to evolve towards the instrument played on the original audio signal. The results obtained in polyphonic piano transcriptions show the feasibility of the approach.

Evolving combat algorithms to control space ships in a 2D space simulation game with co-evolution using genetic programming and decision trees

Developing artificial behaviours to control artificial creatures or vehicles is a task that can be employed by means of Evolutionary Algorithms. A games artificial intelligence is usually developed by seasoned game developers, which need critical knowledge of the games mechanics and rules. This paper presents an alternative, using evolutionary computation to evolve combat algorithms that will allow spaceships to fight effectively in a 2D space simulation game. These combat algorithms will take into account the spaceships characteristics, using them to gain the advantage needed to fight effectively

Cooperative and decomposable approaches on royal road functions: overcoming the random mutation hill-climber

Traditionally, evolutionary algorithms (EAs) encode each individual as a possible solution to the whole problem. As a natural extension to standard EAs, problem decomposition emerged for addressing complex problems. Although many problem decomposition methods rely on dividing the main problem in several less complex sub-problems, launching independent populations (species) to solve each of them, there are also other problem decomposition approaches that require a single population. Parisian approach [1] (often called Individual Evolution) and Gene Fragment Competition (GFC) [5] are two single-population problem decomposition approaches, where the problem can be decomposed in smaller sub-problems, so that they can be evaluated individually reducing the size of the search space. In order to evaluate both Parisian approach and Gene Fragment Competion on problem solving, we aim to use the so called Royal Road functions [3]. We take advantage of the modular and hierarchical structure of the Royal Road test functions to adapt them to both Individual Evolution and Gene Fragment Competition. It is our claim that these functions may serve in the theoretical studies of singlepopulation problem decomposable approaches, such as the Parisian and Gene Fragment Competition, since the landscape can be varied in a number of ways, and the global optimum and all possible fitness values are known in advance. Besides presenting a comparison between a standard genetic algorithm and both Individual Evolution and Gene Fragment Competion on several instances of the Royal Road functions, it is also presented a comparison between these single-population problem decomposition approaches and the results of a previous study made by Ochoa et al. about multi-population co-evolutionary approaches to the same Royal Road functions [4]. One final comparison was

A novel approach to automatic music transcription using electronic synthesis and genetic algorithms

This paper presents a novel approach to the problem of automatic music transcription using electronic synthesis with genetic algorithms. Although the problem is well known and different techniques have been applied before, evolutionary algorithms have never been considered when addressing this problem. We show that, by means of a series of steps, a polyphonic MIDI file -containing instrument’s partiturescan be automatically generated from an audio recording, by extracting and separating simultaneous notes. We describe also the future steps of our research in order to improve the genetic algorithm: increasing performance and decreasing memory usage, extracting the instrument’s features, accurate transcription of note’s duration and, if necessary, the employment of parallel systems. The results obtained shows the feasibility of the approach.

CGP4Matlab - A Cartesian Genetic Programming MATLAB Toolbox for Audio and Image Processing.

This paper presents and describes CGP4Matlab, a powerful toolbox that allows to run Cartesian Genetic Programming within MATLAB. This toolbox is particularly suited for signal processing and image processing problems. The implementation of CGP4Matlab, which can be freely downloaded, is described. Some encouraging results on the problem of pitch estimation of musical piano notes achieved using this toolbox are also presented. Pitch estimation of audio signals is a very hard problem with still no generic and robust solution found. Due to the highly flexibility of CGP4Matlab, we managed to apply a new cartesian genetic programming based approach to the problem of pitch estimation. The obtained results are comparable with the state of the art algorithms.

Evolutionary algorithms and automatic transcription of music

The main problem behind Automatic Transcription (Multiple Fundamental Frequency - F0 - Estimation) relies on its complexity. Harmonic collision and partial overlapping create a frequency lattice that is almost impossible to deconstruct. Although traditional approaches to this problem of rely mainly in Digital Signal Processing (DSP) techniques, evolutionary algorithms have been applied recently to this problem and achieved competitive results. We describe all evolutionary approaches to the problem of automatic music transcription and how some were improved so they could achieve competitive results. Finally, we show how the best evolutionary approach performs on piano transcription, when compared with the state-of-the-art.