|
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.
This process
of reducing the amount information the speech signal called parameterization.
Since the component present only due quantization error offset
it has effect speech recognition, but may have negative impact the used
algorithms which assume signal with zero components.
Furthermore, the signal divided into short frames time which are
processed separately.
Signal processing performed only once for each time frame and, moreover,
is largely accomplished the hardware.1 Division into frames and preprocessing
The basis any speech processing record the signal.
In practise, however, need limit the input signal with band-pass filter or
sample with higher sampling frequency and then apply digital antialiasing filter. For this purpose
. For the purpose speech recognition the division ms
where every frame covers the previous one (i.1 Signal preparation
Before processing with neural network, the signal has processed to
contain only information relevant for recognition. also avoids potential signal
distortion the edges the frame where signal was cut off.1.29
6.e. first frame from 25
ms, second frame from and on). Furthermore, useful adjust the signal into
an appropriate format that the recognizer will able work well with.
The most often used sampling rate according the sampling theorem kHz for
speech signal processing (carrying only human speech), since for most phonemes,
almost all the energy contained the 100 kHz range.
6. This section
includes sampling and quantization audio input which mostly provided by
specialized hardware where the user’s task only setting the sampling frequency. This means that inappropriate or
useless content has removed.1)
where length the frame
sn nth
sample the frame
Finally, the signal suppressed the edges frames that any given
time the most important will the central part.
Otherwise, frequency components that not exist the original signal would be
added the speech signal and distort it. Therefore additional transformation is
applied the final framework which helps the further work with the samples:
averaging and weighting with Hamming window. component is
subtracted from each frame according the formula:
1
0
' N
i
inn s
N
ss (6
