You are on page 1of 16

SLANG DETECTION

USING SPEECH
TEAM MEMBERS:

• Sukavasi.Manasa - 171FA04177
• Batchu.Chandana - 171FA04191
• Eluru.Rajeswari - 171FA04199

Submitted To:

MR.S.ShivaPrasad,

Assistant Professor,

CSE Department.
ABSTRACT:

• With recent advancement in technology voice recognition has become one of the most
efficient measures to provide protection to human’s belongings.

• By this advancement technology we can find whether the speaker was male/female or
which language he/she is speaking or which slang that particular speaker is speaking.

• So, by using this techniques we are going to detect the slang of the speaker whether h
e/she belongs to Telangana/Andhra/Rayalaseema.

• This slag detection is done by using speech. As we all know that when a person is spe
aking then by recognizing their voice we need to detect that particular person belong t
his/that slang.
• So first we need to recognize their voice. Generally, this voice recognition is divide into 2 categori
es :

1. Text Dependent

2. Text Independent

• Text Dependent means it just generally depends on text that what the speaker is speaking about. Th
is technique uses for password security purposes etc.

• And the next technique text independent means it is irrespective of what he/she(speaker) is saying
about.

• So we are going to use this text independent technique in order to find out the slang of the person.

• In deeper of the project we are using MFCC(Mel Frequency Cepstral Coefficient), GMM(Gaussian
Mixture Modelling), EM(Expectation Maximization) .
LITERATURE REVIEW:
• The whole project was mainly divide into three stages:
• Feature extraction
• Training
• Testing
• First stage is Feature extraction, in this feature means the voice features which are tak
en from the person.

• These features are extracted by using MFCC technique. This technique is commonly u
sed as feature extraction in speech recognition.which can be automatically recognize t
he voice of person.

• And next these features are given to the training stage,well we using GMM technique
in order to train all these features.
• And EM technology is also used in GMM in order to find out the mean and variances
that come to know when we go deeper into the project and these features will store in
a database.

• And next stage in testing in this we need to test whether we are getting accurate accur
acy or not.

• In this we just give a dynamic voice , and we probably test whether we are identifying
/ detecting the slag or not.
P ROP O S E D M E T H OD S :

MFCC(Mel Frequency Cepstral Coefficient):


• As I said above this technique is used for feature extraction and the following are the
steps involved in MFCC:
• Pre-emphasizing`
• Framing
• Windowing
• Fast Fourier transform
• Mel filter
• Frequency wrapping
• Discrete cosine transform
GMM(Gaussian Mixture Modelling):
• Before we talking about GMM we need to talk about Clustering. Clustering refers to
grouping similar data points together,based on their attributes and features.

• One of the most popular clustering algorithms is k-means,well k-means means it is a


distance-based algorithm this means that it tries to group the closest points to form a
cluster.

• K-means algorithm seems to be working well, but when we look closely , you will n
otice that all the clusters created have a circular shape because every time it was upd
ated by mean value.

• But when we are considering an example where the distribution of points is not in a
circular form then when we are applying k-means it results too failure.
• So instead of using a distance-based model,we will now use a distributed-based model
.

• So here comes we are using GMM model.

• GMM assume that there are a certain number of Gaussian distributions,and each of th
ese distributions represent a cluster. It tends to group the data points belonging to a si
ngle distribution together.

• Gaussian distributions has a bell-shaped curve,with the data points symmetrically dist
ributed around the mean value.

• For a data set with d features,we would have a mixture of k Gaussian distributions (w
here k is equivalent to the number of clusters)

• We need to find mean and variance matrix so here come another technique called EM.
EM(Expectation-Maximization):
• This algorithm is mainly used to find out the different parameters when the data is mi
ssing/ incomplete.

• This algorithm has 2 steps:


• E-STEP: In this the available data is used to estimate the values of the mis
sing variables.

• M-STEP: Based on the estimated values generated in the E-step , the com
plete data is used to update the parameters.
• Based on the updated values generated from this step , we calculate the new probabilit
ies for each data point and update the values iteratively.So this how the GMM works.
RESULTS:
Noise removal code:
Accuracy:
CONCLUSION:
• So at-last what we conclude is by using this project a machine can detect a persons sl
ag by his/her voice.

• We generally include only three slangs but we can include as many as we want so it m
ay be a future scope for this project.
REFERENCES:
• Some of the references that we took to complete this project are:

• https://www.analyticsvidhya.com/blog/2019/10/gaussian-mixture-models-clustering/

• https://www.ijcsi.org/papers/IJCSI-12-2-126-135.pdf
THANK YOU

You might also like