|
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.
The backpropagation algorithm also
described which one the basic types neural network training. The first half this paper describes the structure and the operation real
and artificial neurons including the description the learning process and the
manner and topology their interconnections. This type network can exceptionally suitable for the language
recognition task too. Both networks were fine-tuned for
optimal functionality.
The same neural network with analysis parameters set into frames ms
and overlaps (NNV2) performed even better.
The goal the work was train the networks with the training words gain
the ability recognizing the language the words and, subsequently, test these
trained networks.
After introducing the Fuzzy Systems and Fuzzy Neural Networks, the Adaptive
Neuro-Fuzzy Inference System (ANFIS) was presented which effectively combines
both neural networks and fuzzy logic reasoning order achieve the best possible
results.
. Both networks were able recognize all the languages. Testing was separated into basic parts: testing the trained network
with words Speaker1 and Speaker2, testing the trained network with words by
Speaker3 (different speaker than train words).
A prerequisite network training acquire training data. The
analysis parameters for the neural network were set into frames where
every frame covers the previous one (NNV1). the framework Matlab,
a language recognition software has been built, which has two different types of
network that can used the ANFIS network and own implementation neural
network trained the backpropagation algorithm. For the training method train folder was
used, which contains 612 words from Speaker1 and Speaker2 (each word times by
both speakers). Fifty-one different words three languages
(English, Czech and Hungarian) were recorded for further network training and
testing purposes for total 153 acquired words (51 English, Czech and 51
Hungarian) three speakers. detailed
insight given into fuzzy systems and fuzzy neural networks including the main
advantages and disadvantages fuzzy systems and the properties both systems
and clearly describes the problems which can solved combining these two
techniques.46
9. our case these
were recordings individual words. CONCLUSION
Within the scope this master´s thesis, tried give deep insight into the
function neural networks, starting with the base the whole concept real
neurons. The network with this setting
precisely identified 124 words out 153 (81,05%) while testing with words by
Speaker1 and 116 words out 153 (75,82%) while testing with words Speaker2. With this setting and 10
cepstral parameters frame the neural network precisely identified 100 words out of
153, which means slightly more than 65% all words (while testing with words by
Speaker1 and Speaker2) and words out 153 (43,14%) while testing with words
by Speaker3. The model Fuzzy Neural Network and Barenji’s ARIC (Approximate
Reasoning Based Intelligent Control) architecture also presented. Every word was recorded times each speaker
which means total amount 1377 words
