|
Kategorie: Diplomové, bakalářské práce |
Tento dokument chci!
Práce popisuje základy principu funkčnosti neuronů a vytvoření umělých neuronových sítí. Je zde důkladně popsána struktura a funkce neuronů a ukázán nejpoužívanější algoritmus pro učení neuronů. Základy fuzzy logiky, včetně jejich výhod a nevýhod, jsou rovněž prezentovány. Detailněji je popsán algoritmus zpětného šíření chyb a adaptivní neuro-fuzzy inferenční systém. Tyto techniky poskytují efektivní způsoby učení neuronových sítí.
e. The number neurons in
each layer also set here.
.m
This function used for audio file loading and parameter calculation (see
Appendix). The function train trains the network for the given
training data. The file names are prepared contain information about the language of
the recording.
Since the average length the recorded words are around 700ms, each file set to
this length cutting the signal this time point and filled with zeros case of
shorter files.m
The signals their raw form are not suitable inputs network because
these contain extremely large amount information. The first letter corresponds the first letter the used languages
(i.
In the case, the mode set 'nnv', the neural network functions created
within the frame Semestral Project MM2E (netinit, netlearn and neteval) are use. Firstly, the file names are determined the given directory that has the
wav extension.
Analysis parameters –params.
Loading audio files wavload. After that, all files are processed sequentially, described herein. The given file read into vector and normalized have maximal
amplitude Subsequently, the parameters are calculated using the params
function. Afterwards, the network tested for
correctness with the same data used for training using the function evalfis. 'c' means Czech, 'e' means English and the prefix 'h' for Hungarian). The number ANFIS networks equals the
number output variables (columns matrix tgt). The input signal limited with band-pass filter with range 100 kHz
to filter out background noise then divided into number frames depending its
length and the adjusted parameters. This
information used for creating the target matrix (tgt) that used for training the
network. For the
case usage neural network, the function feedforwardnet used, which
creates neural network suitable for classification tasks. Further signal preparation described in
Chapter 6. utilizes hybrid learning
technique, what combination the least-squares estimator (LSE) method and
the error backpropagation (EBP) algorithm.1.
These functions can create simple neural network structure, and are able train
and evaluate it. However, parameters can be
used instead the original signals that describe the signal shape appropriate
level. The
result each network saved the corresponding column matrix res.33
The learning itself realized the function anfis that executes the learning
algorithm individually for each network. The target matrix and the matrix parameters are returned return values
of the function
