Papers by Daniel Brojboiu
With advancements in technology and the permanent growth of the music industry there is increasin... more With advancements in technology and the permanent growth of the music industry there is increasing demand for more accurate tools and devices to aid in the production of a musical record. This study aims to explore the use of a deep neural network for blackbox modelling of a hard clipping distortion circuit. The project explores the use of a recurrent neural network that is a stateful long short-term memory system (LSTM) to model the Boss DS-1 distortion guitar pedal. The training dataset consists of a relatively varied selection of samples and play styles and is effective in training the model with under 6 minutes of audio data. Testing the models showed a very good correlation between the target sound and the modelled one. Subjective tests that were also conducted proved that the modelled sound matched very well with the reference, the participating listeners not being able to tell the difference between the original version and the digital one in most cases. Experimental data has also shown that the model performs very well when the pedal is not driven to maximum distortion.
This paper aims to show a possible simplified digital implementation of a well known analog disto... more This paper aims to show a possible simplified digital implementation of a well known analog distortion guitar pedal. The main distortion method is a diode clipper together with a low-pass filter that follows a non linear differential equation that is computationally expensive in real time. The method used is simplified, comprising of cascaded filters to equalise the sound together with a memoryless nonlinearity that is chosen for its efficiency. The design of such algorithms sometimes involves turning the parameters of the digital model to match the sound and waveform to the analog prototype. In this case, the digital implementation of the pedal was done after analysation of analog prototype circuits. Tests and comparisons between the digital implementation, the analog circuit simulation and the real pedal show an overall good reproduction of the actual pedals.
The software presented in this dissertation is designed to modify the listening experience of an ... more The software presented in this dissertation is designed to modify the listening experience of an individual using a playing device, based on measured characteristics of the frequency response of the sound system, as well as any hearing impairment the user might have. The measured characteristics define a personal hearing profile. Using signal processing, certain critical frequencies will be adjusted according to the hearing profile.
One particular concern of this project was the portability of the resulted software to multiple operating systems, such as Mac OS X, Linux or Windows. Python programming language makes it easy to bundle the code into a single folder or application that can run on any of the above mentioned operating systems. The application can be re-written to permit it to run on portable devices using Android or iOS.
Uploads
Papers by Daniel Brojboiu
One particular concern of this project was the portability of the resulted software to multiple operating systems, such as Mac OS X, Linux or Windows. Python programming language makes it easy to bundle the code into a single folder or application that can run on any of the above mentioned operating systems. The application can be re-written to permit it to run on portable devices using Android or iOS.
One particular concern of this project was the portability of the resulted software to multiple operating systems, such as Mac OS X, Linux or Windows. Python programming language makes it easy to bundle the code into a single folder or application that can run on any of the above mentioned operating systems. The application can be re-written to permit it to run on portable devices using Android or iOS.