Neural Networks for Engineering Students
Read the top first. Elon Musk thinks this is more dangerous than nukes. Show a little respect, but don't be afraid.
Scriptie voorgelegd tot het behalen van de graad van Burgerlijk Ingenieur in de Computerwetenschappen: Informatie- en Communicatietechnologie, juni 2006
(Downloaded from LaTeX templates en logo's)
This Project posits elementary analogies of existing Probabilistic and Machine Learning models that have been used to find solutions to the problem of the Structural Segmentation of Musical audio. I have tried to use the idea that the chord of a given beat or frame of a song is an analogous representation of the states generated by trained Hidden Markov Models in generating feature vectors for the aforementioned problem; and that the knowledge of the temporal boundaries within which, a group of frames lie, can be used as constraints in creating the feature vectors that are eventually clustered to identify the pattern in which the various segments of a song repeat.
There are several voice communication systems that are used nowadays which are capable of maintain voice calls between two users in real time. Telephones are widely used all around the world in an unlimited kind of situations. All of these situations expose the microphone (or microphones) of the phones to different and unpredictable noises, as street noise, sea noise, rain noise, wind noise, unwanted voices, car motors, etc. As microphones capture all the sounds around it, including the wanted voice and the unwanted noises, it is necessary to implement digital real time filters capable of attenuate as much as possible all the surrounding noises.
It exists a large quantity of noise reduction methods that have been used in the calling algorithms of phones. Even if these methods have had, in general, a good performance, there is still a research being done in this area in order to improve the current results. Because of this, the multichannel methods were created (using multiple microphones) as well as new algorithms that pretend to have a better noise reduction than the single channel methods. Most of these methods require a speech presence probability (SPP) method to achieve the noise reduction.
The following document presents a research about different SPP methods as well as a comparison between these. This includes an explanation on how theses algorithm work, a Matlab implementation using real voice and noise recordings and objective tests of the filter.
En este documento se muestra un acercamiento identificación de especies arbóreas mediante Histogramas De Gradientes Orientados y maquinas de soporte Vectorial.
Este trabalho tem a missão de explicar alguns temas de sistemas distribuídos, estes
sistemas são fundamentais para a comunicação, integração e processamento de dados via
internet ou rede local. Ao longo do trabalho abordaremos os principais tópicos tais como:
WebServices, Padrões de Sockets entre outros.