Scribd
Upload
52 views4 pages

Report 1

The document reports on a project involving block level and circuit level designing of an isolated word recognition system. At the block level, the system extracts speech features, labels the extracted data, and uses a hidden Markov model to calculate recognition probabilities stored in a database. At the circuit level, an artificial neural network is designed from the database to recognize words by comparing inputs to stored models. MATLAB simulations were performed to implement part of the HMM as hardware and train the neural network.

Uploaded by

Midhun Viswam
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
52 views4 pages

Report 1

The document reports on a project involving block level and circuit level designing of an isolated word recognition system. At the block level, the system extracts speech features, labels the extracted data, and uses a hidden Markov model to calculate recognition probabilities stored in a database. At the circuit level, an artificial neural network is designed from the database to recognize words by comparing inputs to stored models. MATLAB simulations were performed to implement part of the HMM as hardware and train the neural network.

Uploaded by

Midhun Viswam
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

PROJECT REPORT

SUBMITTED BY:-

As a part of the continuation of the main project that we have started in the seventh semester,
we have decided to have a project flow that contain:-

Block level designing

Circuit level designing


On this day, we studied about each blocks in block level designing. The details are
given below.

BLOCK LEVEL DESIGNING

We have analysed each block in the schematic diagram of VQ/HMM isolated word
recognition system as shown in the figure below.

As the input, audio signal is given to the PC. The PC takes this audio signal as wmv
file. This is given to feature extraction block where features of speech is extracted on the
basis of MEL frequency and pitch.

In discrete HMM system, the continuous feature space is subdivided by a vector


quantiser into J non-overlapping subset and each subset is represented within a code word.
The set of available code words is termed the code book. Each data after the feature
extraction is send to labelling block where each extracted datas of speech is labelled. From
feature extraction, by taking the FFT we can take the coefficient of the vectors. By using the
hidden markov model, we can calculate the probability of each models and that is kept as a
library or data base. From this library, input is given to the multiplexer recognised word is
obtained as the output.

On this day, we studied about each blocks in circuit level designing. The details are given
below.

CIRCUIT LEVEL DESIGNING

From the block level designing, the datas labelled and saved in the labelling block are
vector quantised, where the input speech is quantised to different levels and given to the data
base. Thus the first input speech is saved in the data base MODEL 1, second to MODEL 2,
etc upto L models and hence calculated the probability of each models and is saved in it.
Thus a data base of L models is created. Now, when we have to recognise a speech, an input
is given to the feature extraction. Then the extracted speech is labelled and sent to data base
where each data is compared with the models. If the input signal is matched with any of the
models, then it is given to the output as the recognised word.

An artificial neural network is designed from the data base where the probability is
calculated. Next we train the network by using some toolkits in Matlab. The tool we used is
simulink. One of the portion of HMM has been implemented as hardware part and is placed
in the neural network and thus the circuit level designing is complete.

By using the toolkits of Matlab, in the command window we have written the
program and assigned the values for x inputs and y outputs. Next we have to design a neural
network. From the toolkit we get the neural network start where we have four toolkits. They
are:-

Fitting tool

Pattern matching tool

Clustering tool

Time series tool

Next we assigned the values of x and y. Then we went for the network architecture
where the number of hidden networks have to be assigned. Then we trained the network and
thus we have plotted the graph from which we could get the best fit. From the toolkit of the
matlab, we have given an input to an audio device from that to a time scope. Thus we got the
output and plotted in a graph. We gave different values for x and corresponding values
where plotted in the graph.
From the matlab simulink toolkit, to the audio device by using microphone, sound
was given as input and is given to a time scope and thus analysed the signal and the
corresponding plot was obtained.

You might also like