Video Transcript Summarizer using Natural

Language Processing

T Susmitha 22951A33A8
K V Tejdeep 22951A33B7
J Pavithra 22951A3368
Video Transcript Summarizer using Natural Language
Processing

A Project Report
submitted in partial fulfillment of
the
Requirements for the award of the degree of

Bachelor of Technology
in
Computer Science and Information Technology

by

T Susmitha 22951A33A8
K V Tejdeep 22951A33B7
J Pavithra 22951A3368

Department of Computer Science and Information Technology

INSTITUTE OF AERONAUTICAL ENGINEERING

(Autonomous)

Dundigal, Hyderabad–500 043, Telangana

May, 2025

© 2025, T Susmitha, K V Tejdeep, J Pavithra. All rights reserved

 DECLARATION

We Certify that

a. The work contained in this report is original and has been done by us under the guidance of
our supervisor(s).

b. The work has not been submitted to any other Institute for any degree or diploma.

c. We have followed the guidelines provided by the Institute in preparing the report.

d. We have conformed to the norms and guidelines given in the Ethical Code of Conduct of
the Institute.

e. Whenever we have used materials (data, theoretical analysis, figures, and text) from other
sources, we have given due credit to them by citing them in the text of the report and giving
their details in the references. Further, we have taken permission from the copyright owners
of the sources, whenever necessary.

Place: Signature of the Students

Hyderabad Date: T Susmitha 22951A33A8
K V Tejdeep 22951A33B7
J Pavithra 22951A3368

iii
 CERTIFICATE

This is to certify that the project report entitled Video Transcript Summarizer
using Natural Language Processing submitted by T Susmitha, K V Tejdeep
and J Pavithra, to the Institute of Aeronautical Engineering, Hyderabad in partial
fulfilment of the requirements for the award of the Degree Bachelor of
Technology in Computer Science and Information Technology is a Bonafide
record of work carried out by them under my/our guidance and supervision. The
contents of this report, in full or in parts, have not been submitted to any other
Institute for the award of any Degree.

Supervisor Head of the Department

Mr. B Siva Sankar Dr. M Purushotham Reddy

Date:

iv
 ACKNOWLEDGEMENT

The satisfaction that accompanies the partial completion of our task would be incomplete without
acknowledging those who made it possible. Their constant guidance and engagement have
crowned all our efforts with success. We extend our heartfelt thanks to our college management
and to the respected Sri M. Rajashekar Reddy, Chairman of the Institute of Aeronautical
Engineering, Dundigal, for providing us with the necessary infrastructure to carry out this project
work.

We are also deeply grateful to Dr. L. V. Narasimha Prasad, Principal, whose invaluable
knowledge and guidance have been instrumental in our work, and to Dr. M. Purushotham
Reddy, Professor and Head of the Department of Computer Science and Information Technology,
for his unwavering support. Our sincere thanks also go to our beloved Principal for their
continuous encouragement.

We are deeply grateful to our Project Supervisor, Mr. B. Siva Sankar, Associate Professor,
Department of Computer Science and Information Technology, and special thanks go to our
Project Coordinator, Ms. M. Himabindu, Assistant Professor, Department of Computer Science
and Information Technology, for their internal support and professionalism in helping us shape this
project into a successful one.

Finally, we take this opportunity to express our thanks to everyone who directly or indirectly
helped us bring this effort to its present form.

T Susmitha 22951A33A8
K V Tejdeep 22951A33B7
J Pavithra 22951A3368

v
 APPROVAL SHEET

This project report entitled Video Transcript Summarizer using Natural

Language Processing by T Susmitha, K V Tejdeep and J Pavithra is approved
for the award of the Degree Bachelor of Technology in Computer Science and
Information Technology.

Examiners Supervisor

Mr B Siva Sankar

Principal

vi
 ABSTRACT

Video content is a widely consumed form of media, often containing lengthy and
detailed spoken information. Manually transcribing and summarizing these transcripts
can be time-consuming and inefficient. This project presents a Video Transcript
Summarizer using Natural Language Processing (NLP) and Python, which automates
the process of converting spoken video content into concise, meaningful summaries.
The system employs speech-to-text APIs to transcribe video audio into text, followed by
preprocessing steps such as tokenization, stop word removal, and lemmatization. Then,
advanced NLP techniques like TextRank and transformer-based models are applied to
extract key information and generate accurate summaries. The summarized output
provides users with the essence of the video content, improving content accessibility
and saving time. This tool has potential applications in education, media analytics, e-
learning platforms, and accessibility tools for the hearing impaired.

In addition to its core functionality, the system can be integrated into video platforms
and dashboards for real-time summarization. It reduces cognitive overload by
presenting only relevant insights, especially in long-form academic or technical videos.
Future enhancements may include multilingual support and summarization
customization based on user preferences. Overall, the project demonstrates the practical
synergy between NLP and video analysis, contributing meaningfully to the field of
intelligent content processing.

Keywords: Video Summarization, Natural Language Processing, Python, Speech-to-

Text,
TextRank, Transformers, Content Accessibility

vii
 CONTENTS

Title Page i
Cover Page ii
Declaration iii
Certificate by the Supervisor iv
Acknowledgement v
Approval Sheet vi
Abstract vii
Contents viii
List of Figures x
Abbreviations xi
Chapter 1 Introduction 1
1.1 Background 1
1.2 Problem Statement 2
1.3 Existing System 2
1.3.1 Demerits of Existing System 3
1.4 Proposed System 3
1.5 Objectives 4
Chapter 2 Review of Relevant Literature 5
2.1 Literature Review 5
2.2 System Study 9
2.2.1 Feasibility Study 9
2.2.2 Economical Feasibility 9
2.2.3 Technical Feasibility 9
2.2.4 Social Feasibility 9
Chapter 3 System Design 10
3.1 System Architecture 10

viii
 3.2 System Description 12
3.3 UML Diagrams 13
3.3.1 Class Diagram 14
3.3.2 Sequence Diagram 15
3.3.3 State Chart Diagram 16
Chapter 4 Methodology and Implementation 17
4.1 Methodology 17
4.2 Components of Video Transcript Summarizer 19
4.3 Architecture of Video Transcript Summarizer 20
4.4 Algorithms 22
4.5 Implementation 23
4.5.1 Hardware and Software Requirements 25
Chapter 5 Testing 26
5.1 Testing Strategies 26
5.1.1 Unit Testing 26
5.1.2 Integration Testing 27
5.1.3 Functional Testing 27
5.1.4 System Testing 27
5.1.5 White Box Testing 28
5.1.6 Black Box Testing 28
Chapter 6 Results and Discussion 29
Chapter 7 Conclusion 31
References 32

ix
 LIST OF FIGURES

Figure Figure Name Page

No Number
3.1 Basic view of system architecture 11

3.2 Class Diagram 14

3.3 Sequence Diagram 15

3.4 State Chart Diagram 16

4.1 Architecture of Video Transcript Summarizer 21

4.2 CNN Layers Visualization 23

4.3 Layered architecture of a CNN 24

4.4 Implementation in Google Colab 25

6.1 Heatmap Visualization of Data 29

6.2 Real vs. Predicted Text Comparison with 29

Accuracy
6.3 Average Accuracy Calculation Result 29

x
 LIST OF ABBRIVATIONS

NLP Natural Language Processing

API Application Programming Interface
OCR Optical Character Recognition
LSA Latent Semantic Analysis
NLTK Natural Language Toolkit
GPU Graphics Processing Unit

xi
 CHAPTER 1

INTRODUCTION

1.1 Background

With the rapid growth of digital media, video content has become a dominant form of
communication, education, and entertainment. From online classes, webinars, and business
meetings to news broadcasts, tutorials, and vlogs, the sheer volume of audiovisual content
being created and consumed daily is staggering. While video content is rich in information, it
often presents challenges for users who wish to quickly grasp the core message without
watching entire recordings. This growing demand for content efficiency has led to the need for
intelligent systems capable of summarizing video content into concise and informative text.

A Video Transcript Summarizer using Natural Language Processing (NLP) aims to solve
this problem by automatically generating textual summaries from spoken content in videos.
The process typically involves two major stages: Automatic Speech Recognition (ASR) and
text summarization. ASR technologies convert the audio stream of a video into machine-
readable text. These transcriptions are then processed using advanced NLP techniques such as
tokenization, part-of-speech tagging, named entity recognition, and sentence boundary
detection to prepare them for summarization. The summarization can be extractive—where
important sentences are selected verbatim—or abstractive—where new sentences are generated
that capture the original meaning more concisely.

This project is particularly useful in a wide range of applications. In education, students can
use transcript summaries to review lectures quickly. In the corporate world, professionals can
summarize long meetings for quick reference. In journalism, summaries can help reporters
pull key insights from video interviews. Additionally, the system supports accessibility,
allowing individuals with hearing impairments to comprehend video content in written form,
and it facilitates search engine indexing, making video content more searchable and
discoverable.

The project combines state-of-the-art machine learning, speech processing, and language
modeling techniques, highlighting how artificial intelligence can be harnessed to improve
productivity, accessibility, and user experience in the digital information age. Through this
work, we contribute to the evolving field of multimodal data processing, where visual, audio,
and textual elements come together to create intelligent, automated systems for understanding
and managing large volumes of unstructured data.

1
 1.2 Problem Statement

In today’s digital world, vast amounts of video content are generated daily across domains like
education, business, and media. However, extracting useful information from lengthy videos
remains a time-consuming and inefficient process. Manually transcribing and summarizing
videos is not only labor-intensive but also impractical at scale.

While Automatic Speech Recognition (ASR) systems have improved the ability to convert
speech into text, they often lack the capability to generate coherent and concise summaries.
Furthermore, challenges such as background noise, speaker variability, and domain-specific
vocabulary can affect transcription accuracy.

There is a growing need for an intelligent, automated system that can effectively convert video
speech into text and generate relevant summaries using Natural Language Processing (NLP).
Such a solution would significantly improve accessibility, information retrieval, and user
productivity across various fields.

1.3 Existing System

Current systems for processing video content primarily focus on converting speech into text using
Automatic Speech Recognition (ASR) tools such as Google Speech-to-Text, IBM Watson, or Amazon
Transcribe. While these services are effective in transcribing spoken words, they do not provide
summarization capabilities, leaving users with lengthy transcripts that still require manual review. In
many workflows, users depend on manual or semi-automated methods to extract key points from the
text, which is time-consuming and not scalable. Furthermore, these systems often struggle with real-
world challenges like background noise, speaker accents, overlapping speech, or domain-specific
vocabulary. They also lack contextual understanding, which limits their ability to produce concise and
coherent summaries. As a result, existing solutions fail to fully address the growing demand for
automated tools that can both transcribe and intelligently summarize video content.

1.3.1 Demerits in Existing system

2
Despite the availability of advanced speech recognition tools, current systems for processing video
content are limited in functionality and effectiveness. Most existing solutions focus solely on
converting speech to plain text using Automatic Speech Recognition (ASR), without providing any
mechanism for summarizing the resulting transcripts. This leaves users with long, unstructured blocks
of text that are difficult to navigate and time-consuming to read. Furthermore, these systems often lack
contextual understanding and semantic analysis, making them incapable of identifying and extracting
key points or summarizing information meaningfully. As a result, users must manually review the
transcript to identify relevant content, which reduces efficiency and increases cognitive load.

Another major drawback is the inability of existing systems to handle complex audio
environments. Factors such as background noise, overlapping speech, varied speaker accents,
and informal language often degrade transcription accuracy. In addition, these systems may
perform poorly in real-time scenarios or with low-quality audio input, further limiting their
reliability. Moreover, many commercial ASR platforms are cloud-based, raising concerns over
data privacy and confidentiality when processing sensitive or proprietary video content.

In educational, corporate, and media settings—where timely access to summarized information

is crucial—these limitations prove to be significant obstacles. The lack of integrated Natural
Language Processing (NLP) capabilities in these systems prevents them from identifying the
intent, tone, and context of the conversation, which are essential for generating coherent and
user-friendly summaries. This demonstrates the pressing need for an end-to-end intelligent
system that combines accurate transcription with automated summarization, providing users with
clear, concise, and contextually relevant outputs.

1.4 Proposed system

The proposed system is an intelligent Video Transcript Summarizer that leverages Natural
Language Processing (NLP) techniques to automatically generate concise and coherent
summaries from video content. It is designed to address the limitations of manual transcription
and summarization by automating the entire workflow—from extracting speech to generating
meaningful text summaries.

2 The system begins by using Automatic Speech Recognition (ASR) technologies (such as Google
Speech-to-Text or similar APIs) to convert the spoken audio in video files into raw text transcripts.
This transcript data often contains noise and inconsistencies, so it is cleaned and refined through a
text preprocessing pipeline, which includes steps like tokenization, stop-word removal,
lemmatization, and sentence segmentation.
3 Once the text is cleaned, the system applies NLP-based summarization techniques. Both
extractive summarization (using algorithms like TextRank, which selects important sentences
from the original text) and abstractive summarization (using models like BART or T5, which

3
 generate new sentences that capture the meaning of the content) can be implemented based on the
use case.
4 The final output is a brief yet comprehensive summary that captures the key points of the
original video content. This makes it easier for users to understand lengthy videos without
watching them in full. The system is scalable, efficient, and applicable in domains such as e-
learning, corporate meetings, media analysis, and assistive technology for the hearing impaired.
It can also be integrated into dashboards or video platforms for real-time summarization.

4
 1.5 Objectives

The objective of this project is to design and develop an automated system that can efficiently
transcribe and summarize video content using Natural Language Processing (NLP) techniques. The
system aims to convert spoken words from video files into accurate text using speech-to-text APIs and
then process this text through a series of NLP methods including tokenization, stop-word removal, and
lemmatization. Once the raw transcript is cleaned, the system employs extractive and abstractive
summarization algorithms to generate concise, meaningful summaries. This enhances content
accessibility for users with hearing impairments and allows users to quickly grasp the core message of
lengthy videos without watching them in full. The project also seeks to improve productivity by
reducing the time required for manual review of video content and aims to offer scalability for
integration with educational platforms, media dashboards, and other real-world applications. In the
future, the system can be extended to support multilingual processing and customizable summarization
outputs based on user preferences.

5
 CHAPTER 2

REVIEW OF RELAVENT LITERATURE

2.1 Literature Survey

This chapter provides an in-depth review of the advancements in lip-reading technology,

with a particular emphasis on the LipNet system and its contributions to the field of visual speech
recognition. The literature survey explores various approaches and techniques developed to
enhance the accuracy and efficiency of lip-reading systems. The survey also evaluates the
performance and impact of these advancements in improving communication accessibility,
offering insights into the current state of the art and identifying areas for future research and
development.

1. Jayesh Chavan et al. [11] developed Transcriber Sum, a deep learning-based tool for
summarizing YouTube video transcripts. Their system employs advanced machine learning techniques
to automatically generate concise summaries, enabling users to quickly grasp key content without
watching the entire video. This approach enhances accessibility and efficiency in content consumption.
2. P. Vijaya Kumari et al. [12] introduced a YouTube Transcript Summarizer utilizing Flask and
NLP. The system extracts transcripts from YouTube videos and applies NLP techniques to generate
summaries. It features a user-friendly interface allowing users to download summaries and share them
via email or messaging platforms, streamlining the information retrieval process.
3. Anusuya A. et al. [13] proposed a Multilingual YouTube Transcript Summarizer that leverages
machine learning and NLP to provide automated summaries of video transcripts in multiple languages.
The system analyzes language patterns and employs algorithms to generate concise summaries, aiding
users in quickly identifying key points across diverse linguistic content.
4. Ilampiray P. et al. [14] presented a Video Transcript Summarizer that utilizes NLP processing
for text extraction and BERT-based summarization. This approach provides users with a textual
description and an abstractive summary of video content, facilitating the discrimination between
relevant and irrelevant information according to user needs.
5. Tengchao Lv et al. [15] introduced VT-SSum, a benchmark dataset for video transcript
segmentation and summarization. The dataset comprises 125K transcript-summary pairs from 9,616
videos, offering a substantial resource for training and evaluating models in the domain of video
transcript summarization.

6
1. Rochan et al. [16] proposed a method for video summarization using fully convolutional sequence
networks. By formulating video summarization as a sequence labeling problem, they adapted semantic
segmentation networks to identify key frames, demonstrating improved performance over recurrent
models.
2. Ji et al. [17] introduced an attention-based encoder-decoder framework for supervised video
summarization. Utilizing Bidirectional LSTM and attention mechanisms, their model effectively
selects keyshots, closely mimicking human summarization patterns.
3. Narasimhan et al. [18] developed CLIP-It!, a language-guided multimodal transformer for both
generic and query-focused video summarization. By leveraging natural language queries and dense
video captions, their model outperforms existing methods in various settings.
4. Zhao et al. [19] presented a hierarchical recurrent neural network (H-RNN) for video
summarization. Their two-layer architecture captures long-term temporal dependencies, making it
suitable for summarizing longer videos with reduced computational complexity.
5. Solanki [20] explored automating video summarization using OpenAI's Whisper for transcription
and Hugging Face's SAMSum model for summarization. This approach highlights the effectiveness of
combining ASR and NLP models for conversational video summarization.
6. Insight7 [21] discussed AI-driven techniques for summarizing video transcripts, emphasizing
methods like keyword extraction, thematic analysis, and significant sentence extraction to distill
essential information from lengthy conversations.
7. Analytics Vidhya [22] demonstrated video summarization using OpenAI's Whisper model for
transcription and Hugging Face's Chat API for summarization. Their approach showcases the practical
application of these tools in various domains, including education and surveillance.
8. Porwal et al. [23] focused on video transcription and summarization using NLP techniques. Their
study underscores the importance of preprocessing and the application of NLP algorithms to generate
concise summaries from video content.
9. Bedi et al. [24] conducted a survey on text and video summarization using machine learning. They
analyzed various deep learning techniques, including LSTM and CNN frameworks, highlighting their
effectiveness in summarizing biomedical transcripts.
10. Wikipedia [25] provides an overview of automatic summarization techniques, discussing both
extractive and abstractive methods. The article emphasizes the role of NLP in generating concise
summaries from large text corpora, including video transcripts.

7
 2.2 System Study
2.2.1 Feasibility Study

2.2.2 The feasibility study is an essential phase in the project development lifecycle, aiming to assess
the viability of the proposed system across various dimensions. For the Video Transcript
Summarizer using NLP, the feasibility analysis evaluates technical, economic, and social
aspects to ensure the project's practicality, effectiveness, and sustainability.

2.2.3 Economical Feasibility

3. From a cost-benefit perspective, the system is economically viable. Most of the required tools
and frameworks are open-source, which minimizes software licensing costs. Moreover, the
project can be developed using existing infrastructure such as personal computers or cloud-
based platforms like Google Colab, reducing the need for additional investments. The
anticipated benefits, such as time savings and enhanced content accessibility, justify the low
development costs.

2.2.4 Technical Feasibility

2.2.5 The system utilizes well-established technologies such as Python, Natural Language Toolkit
(NLTK), Speech-to-Text APIs, and transformer-based models (e.g., BART, T5). These tools
are widely supported, open-source, and compatible with current computing environments.
Additionally, the required computational resources, such as Google Colab or local machines
with moderate hardware (e.g., 4GB RAM and a standard GPU), are sufficient for both
development and testing. Therefore, the technical implementation is considered feasible.

2.2.6 Social Feasibility

The system is designed to improve accessibility and usability, particularly for students, professionals,
and individuals with hearing impairments. By providing concise summaries of video content, it
reduces information overload and supports efficient knowledge acquisition. The system is user-
friendly, non-intrusive, and does not require specialized training, making it socially acceptable and
beneficial to a broad user base.

8
 CHAPTER 3 SYSTEM

DESIGN

3.1 System Architecture

The architecture of the Video Transcript Summarizer using Natural Language Processing
(NLP) follows a modular pipeline designed to extract, process, and summarize spoken content
from video files. This system is divided into several interconnected modules, each responsible for
a critical stage in the summarization process, ensuring efficient data flow and accurate output
generation.

Video Input Module

The system begins with the Video Input Module, where users upload video files containing
spoken content. This module extracts the audio stream from the video using multimedia processing
tools such as moviepy or ffmpeg. The extracted audio is then forwarded to the Speech-to-Text
(STT) Module, which uses external APIs or pre-trained models like Google Speech-to-Text,
Whisper, or IBM Watson to convert the audio into a raw text transcript.

Text Preprocessing Module

The transcript is passed into the Text Preprocessing Module, which performs essential
Natural Language Processing tasks. These include tokenization (breaking text into words or
sentences), stop-word removal (filtering out common but irrelevant words), lemmatization
(reducing words to their root forms), and punctuation correction. This cleansed text is then
forwarded to the summarization pipeline.

Summarization Module

The Summarization Module is responsible for generating the condensed version of the
transcript. It supports two approaches: Extractive summarization, which uses algorithms like
TextRank to select the most important sentences, and Abstractive summarization, which
employs transformer-based deep learning models such as BART or T5 to generate context-aware,
rephrased summaries that retain the original meaning.
1. Extractive summarization, which uses graph-based algorithms like TextRank to identify
and select the most relevant sentences directly from the transcript.
2. Abstractive summarization, which employs transformer-based models such as BART or T5
to understand the context and generate new, concise summaries using rephrased text.

9
Display Module

The final output is managed by the Display Module, which presents the summarized
content to the user in a readable format via a web or desktop interface. The system also supports
optional features such as exporting the summary to text files or integrating it into dashboards for
extended accessibility. This layered architecture ensures modularity and scalability, allowing
each component to be improved or replaced independently. It also enhances usability and
accuracy, enabling the system to deliver fast and precise video transcript summaries that improve
content accessibility, especially for educational, professional, and assistive applications.

Conclusion

The CNN-LipNet architecture outlined above offers a robust framework for lip reading. By
effectively combining the strengths of CNNs for spatial feature extraction and LSTMs for
temporal modelling(as shown in figure 3.1), this system demonstrates exceptional capabilities in
recognizing spoken words from lip movements in video sequences.

Figure 3.1: Basic view of system architecture

10
3.2 System Description

The Video Transcript Summarizer using Natural Language Processing (NLP) is

designed to automate the process of extracting key information from video content by converting
speech into text and generating concise summaries. The system is structured as a multi-stage
pipeline, with each module performing a specific and essential function to ensure accurate and
meaningful summarization of video transcripts.
The process begins with the Video Input Module, which enables users to upload videos in
common formats such as MP4 or AVI. Once a video is uploaded, the system uses multimedia
processing libraries to extract the audio stream from the file. This step isolates the spoken content
and prepares it for transcription.
The extracted audio is processed by the Speech-to-Text (STT) Module, which employs
advanced speech recognition APIs like Google Speech-to-Text or OpenAI Whisper to generate a
raw text transcript. This module is capable of handling different accents, background noise, and
varying speech rates, although the accuracy may depend on the quality of the audio input.
The transcript is then passed to the Text Preprocessing Module, where it undergoes a series
of cleaning and normalization steps. These include removing filler words and noise, correcting
punctuation, and applying tokenization, stop-word removal, and lemmatization. This ensures that
the input to the summarization engine is grammatically sound and semantically clean.
Following preprocessing, the system moves to the Summarization Module, which is the
core of the architecture. It supports both extractive and abstractive summarization techniques.
Extractive summarization involves selecting key sentences directly from the transcript using
algorithms like TextRank. Abstractive summarization, on the other hand, leverages deep learning
models such as BART or T5 to generate new, shortened text that captures the essence of the
original input in a more fluent and human-like manner.
Finally, the processed summary is sent to the Display and Output Module, which renders
the output in a user-friendly format. Users are provided options to view the summary, copy it, or
export it as a text or PDF file. The modular design of the system allows easy integration with
web dashboards, educational tools, or accessibility platforms, making it highly adaptable for real-
world use.
Overall, the system provides a reliable and efficient solution for automatic video
summarization, reducing cognitive load and improving accessibility, especially in contexts such
as online education, meeting transcription, content management, and assistive technologies.

11
3.3 UML Diagrams

A UML diagram is a way to visualize systems and software using Unified Modelling
Language (UML). Software engineers fabricate UML diagrams so that they could understand
how systems are designed, what kind of code architecture and proposed implementation are
inside the software. The flow diagrams and activity workflows can be modelled using the same
UML diagrams as well. Coding might be pretty difficult because the elements here are
interconnected. There are usually hundreds or even thousands of code lines that can appear too
cluttered up to a human eye. This diagram represents the information by a neat visual picture that
is more digestible. It employs a common practice for the system specification approved
graphically to give a shape to said conceptual thoughts.

The main objectives of designing UML diagrams are:

 The primary goal of UML is to define some general-purpose simple modelling language so
that all modelers can use and understand.

 UML is not a development method rather it accompanies with processes to make a

successful system.

 UML diagrams are the representation of object-oriented concepts only. Thus, before
learning UML, it becomes important to understand OO concept in detail. ∙ UML can be
described as the successor of object-oriented analysis and design. ∙ Provide mechanisms for
expansion and specialized core concepts.

12
3.3.1 Class Diagram

A class diagram is a key component in object-oriented modeling, used to represent the static
structure of a system by detailing its classes, attributes, methods, and relationships. For the Video
Transcript Summarizer using NLP, the class diagram illustrates how the main components of
the system interact, reflecting the core functionalities and data flow between different modules.
The central class in the diagram is the VideoTranscriptSummarizer class, which acts as
the controller and coordinates interactions between all other classes. It is responsible for initiating
the summarization process and managing data flow between modules.

Figure 3.2: Class Diagram

13
3.3.2 Sequence Diagram

The sequence diagram illustrates the dynamic interaction among the principal objects
involved in generating a summary from an uploaded video. It captures the chronological flow of
messages exchanged from the moment a user submits a file until the system returns the final
condensed text.
1. User initiates the process by invoking uploadVideo() on VideoInput.
2. VideoInput verifies the file format with validateFormat() and, upon success, calls
extractAudio() on AudioExtractor.
3. AudioExtractor isolates the soundtrack and forwards the resulting audio path to
SpeechToTextEngine via convertAudioToText().
4. SpeechToTextEngine performs automatic speech recognition, returning the raw transcript
to TextPreprocessor through cleanTranscript().
5. TextPreprocessor executes a pipeline of tokenize(), removeStopWords(), and lemmatize(),
then passes the refined text to Summarizer.
6. Summarizer decides—based on configuration—whether to invoke runTextRank()
(extractive) or runAbstractiveModel() (transformer) and produces the concise summary.
7. The summary is delivered to OutputHandler, which triggers displaySummary() for on-
screen viewing and optionally exportToFile() for download.
8. OutputHandler returns a confirmation to the User that the summary is ready.

Figure 3.3: Sequence Diagram

14
3.3.3 State Chart Diagram

The state chart diagram models the various states the system transitions through during the
video summarization process. It describes the lifecycle of the system from the point a video is
uploaded to when a summarized output is delivered to the user. Each state represents a specific
phase of system activity, while the transitions indicate the conditions or events that trigger
movement from one state to another.
The system begins in the Idle State, where it awaits user interaction. Upon uploading a valid
video file, the system transitions to the Video Uploaded state. It then enters the Audio
Extraction state, where the video file is processed to isolate the audio stream. Once audio is
successfully extracted, the system moves into the Speech-to-Text Conversion state, in which the
audio content is converted into a textual transcript using a speech recognition engine.
After transcription, the system transitions into the Text Preprocessing state. Here, natural
language processing techniques such as tokenization, stop-word removal, and lemmatization are
applied to clean the transcript. Upon successful preprocessing, the system enters the
Summarization state, where either extractive or abstractive methods are used to generate a
summary.
Once the summary is generated, the system reaches the Summary Ready state. In this state,
the user can view, copy, or download the summary. The process concludes when the user exits
the session or initiates a new summarization task, at which point the system returns to the Idle
State.
This state chart ensures that the system maintains a clear and logical flow, transitioning
smoothly between tasks while handling input and output effectively. It also helps in managing
errors—for instance, if audio extraction or transcription fails, the system can revert to the
previous stable state for user correction.

15
Figure 3.4: State Chart Diagram

16
 CHAPTER 4

METHODOLOGY AND IMPLEMENTATION

4.1 Methodology

The development methodology for the Video Transcript Summarizer follows a systematic,
modular, and iterative approach. Each phase of the system is implemented and tested
independently before full integration. The system is divided into well-defined stages, each
responsible for a specific function in the overall summarization pipeline. The methodology
focuses on efficient data flow, modular design, and the use of advanced Natural Language
Processing and Speech Recognition technologies.
The process begins with the video upload, where users submit a video file to the system.
The audio extraction module processes the video and isolates the audio stream using tools like
moviepy or ffmpeg. Once the audio is extracted, it is passed to a Speech-to-Text (STT) engine,
such as Google Speech-to-Text or OpenAI Whisper, to generate the raw transcript from the
spoken content in the video.
Following transcription, the text preprocessing phase is applied to clean and prepare the
data for summarization. This includes tokenization, removal of stop words, lemmatization, and
sentence segmentation. These steps ensure the input to the summarizer is semantically structured
and free from unnecessary noise.
After preprocessing, the cleaned transcript is sent to the Summarization module, where the
core logic of the system resides. Depending on the design choice, either extractive
summarization (using graph-based algorithms like TextRank) or abstractive summarization
(using deep learning models like BART or T5) is employed to generate a concise, meaningful
summary of the video content.
The final stage is the output module, where the generated summary is presented to the user.
Options such as viewing, downloading, copying, or sharing the summary are offered through a
simple and intuitive interface. The entire methodology is designed for scalability and user
accessibility, and allows for future integration with multilingual support, keyword extraction, or
even real-time summarization pipelines.

17
4.2 Components of Video Transcript Summarizer
The implementation of the Video Transcript Summarizer using NLP consists of several
essential components, each of which contributes to the functionality of the system in a modular
and coordinated manner. These components work sequentially, transforming raw video content
into a clean, concise textual summary that enhances information accessibility and usability.

4.2.1 Video Input Module

This module is responsible for receiving and validating the video file uploaded by the user. It
ensures that the file format is supported (e.g., .mp4, .avi) and forwards the video to the next stage
for audio extraction.

4.2.2 Audio Extraction Module

Using libraries like moviepy or ffmpeg, this component isolates the audio stream from the video
input. The extracted audio is converted to a .wav format, which is optimal for speech-to-text
conversion models.

4.2.3 Speech-to-Text (STT) Module

This is a critical component that converts the extracted audio into a raw transcript. The module
can utilize APIs like Google Speech-to-Text, OpenAI Whisper, or other pretrained Automatic
Speech Recognition (ASR) models. The output is unstructured text that includes all spoken
words.

4.2.4 Text Preprocessing Module

This module cleans and prepares the transcript for summarization. It performs:

 Tokenization: Splitting text into individual words or sentences.

 Stop-word Removal: Filtering out commonly used but irrelevant words.

 Lemmatization: Reducing words to their root forms.

 Punctuation & Case Normalization: Standardizing grammar for consistency.

4.2.5 Summarization Module

This is the core processing unit where meaningful summaries are generated. It supports:

18
  Extractive Summarization: Selects key sentences using algorithms like TextRank.

 Abstractive Summarization: Uses transformer models like BART or T5 to rephrase and

shorten the content while preserving meaning.

4.2.6 Output Module

The final module is responsible for presenting the summary to the user in a clean and readable
format. It supports options for displaying the summary, copying it, exporting it as .txt or .pdf,
and even sharing it through connected applications.

19
4.3 Architecture of Video Transcript Summarizer

The architecture of the Video Transcript Summarizer consists of several integrated

modules working in sequence. It starts with the Video Input Module, where a YouTube video
URL or video file is provided. The Audio Extraction Layer uses tools like yt-dlp to download
and extract audio from the video. This is followed by the Speech-to-Text Module, which
transcribes the audio into text using APIs like Whisper, Google Speech-to-Text, or
YouTubeTranscriptApi. The resulting raw transcript is then passed to the Text Preprocessing
Module, which performs cleaning operations such as removing noise, stopwords, and performing
lemmatization. The cleaned text is then fed into the Summarization Engine, which applies
either extractive (e.g., TextRank) or abstractive (e.g., T5, BART) NLP models to generate
concise summaries. Optionally, a Language Detection and Translation Module ensures
support for regional languages like Telugu and Hindi. Finally, the output is rendered through the
User Interface Layer, typically built with Streamlit or Flask, allowing users to view and
download the summarized content in a clean and accessible format.

Figure 4.1: Architecture of Video Transcript Summarizer

20
Video Input Module:
 This module serves as the entry point of the system where users provide a video source. It
typically accepts a YouTube URL or a locally stored video file. The module ensures the
validity of the input and initiates the pipeline by passing the video to the next stage. It handles
basic input validation, error handling, and user interface interactions if a front-end is involved.
This module is crucial for starting the summarization workflow in a user-friendly and
accessible way.

Audio Extraction Layer:

 Once the video is received, the audio extraction layer isolates the audio track from the video
file. This is commonly done using tools like yt-dlp or ffmpeg, which allow precise control
over the format and quality of the extracted audio. The extracted audio is typically saved in a
format like .wav or .mp3 for compatibility with transcription services. This module ensures
high-quality audio output, which is essential for accurate transcription..

Speech-to-Text Module:

 The core of the system’s transcription process, this module converts spoken audio into textual
transcripts. It uses speech recognition engines such as OpenAI’s Whisper, Google Cloud
Speech-to-Text, or YouTubeTranscriptApi for YouTube videos with available captions. The
choice of engine may vary based on factors like language support, accuracy, and cost. This
module is designed to work seamlessly with multiple languages and dialects, ensuring robust
transcription even for regional content.

Text Preprocessing Module:

 This module prepares the raw transcript for summarization by cleaning and structuring the text.
It performs tasks like tokenization, stopword removal, punctuation cleanup, and lemmatization.
Sentence segmentation is also applied to break the transcript into meaningful units. These
preprocessing steps are vital to reduce noise and improve the quality of the summary generated
by downstream NLP models.

Summarization Engine:

 The summarization engine is the heart of the project, responsible for generating a concise and
informative summary from the preprocessed text. It can use extractive methods like TextRank
to select key sentences or abstractive methods like BART, T5, or GPT to generate new,
coherent sentences that convey the main ideas. The choice between extractive and abstractive
summarization depends on the use case and resource availability. This module ensures that the
final output is readable, relevant, and accurate.

21
Language Detection and Translation Module:
 To support multilingual videos, this module detects the language of the transcript and, if
necessary, translates it into a target language like English. It uses libraries like langdetect or
services like Google Translate API to provide this functionality. For regional language videos
such as Telugu or Hindi, this module ensures the summarizer can operate effectively even
when the input is not in English, broadening the system’s usability.

User Interface Layer:

 The final module is the user-facing component, often built using frameworks like Streamlit or
Flask. It allows users to input video links, select languages, view the generated transcript and
summary, and download results. This module focuses on providing a simple, clean, and
interactive experience, making the tool accessible to both technical and non-technical users.

22
 4.4 Algorithms

Speech Recognition Algorithm: Whisper (by OpenAI):

 Converts spoken audio to text.

 Whisper is a deep learning-based automatic speech recognition (ASR) system trained on a
large multilingual dataset. It uses a transformer encoder-decoder architecture to transcribe
speech, handling background noise and different accents well.

Text Preprocessing Algorithms:

 Tokenization : Splits the text into words or sentences using NLP libraries like NLTK or
spaCy.
 Stopword Removal: Removes common words (like "is", "the", etc.) that add little meaning.
 Lemmatization/Stemming: Reduces words to their root forms for consistency.
 Use in project: Cleans and normalizes the transcript before applying summarization models.

TextRank Algorithm (Extractive Summarization):

 Represents sentences as nodes in a graph.

 Connects nodes based on similarity (cosine similarity of TF-IDF or word embeddings).
 Ranks sentences using a variant of the PageRank algorithm.
 Selects top-ranked sentences as the summary.

T5 / BART (Abstractive Summarization):

 Pretrained transformer models fine-tuned on summarization datasets (like CNN/DailyMail).

 Reads full text and produces a shorter version using learned language generation patter

23
Language Detection and Translation:

 Uses models like langdetect for detection and Google Translate or MarianMT for translation.

 Enables multilingual support (e.g., Telugu, Hindi), making the system accessible to non-
English users.

4.5 Implementation

Figure 4.2: Implementation

The implementation of the Video Transcript Summarizer project involves a structured pipeline built
using Python, integrating components of NLP, machine learning, and audio processing.

The process begins by accepting a YouTube video URL or a local video file, from which audio is
extracted using tools like yt-dlp or ffmpeg. This audio is then transcribed into text using an automatic
speech recognition (ASR) model such as OpenAI’s Whisper, which efficiently handles multilingual
input and varying accents. The raw transcript undergoes preprocessing, including tokenization,
stopword removal, and lemmatization, to clean and structure the text. For summarization, the system
offers two approaches: extractive summarization using algorithms like TextRank, which selects key
sentences based on their semantic importance, and abstractive summarization using models like T5
or BART, which generate a condensed version of the content in natural language. Optionally, the
system can include a language detection and translation module to support regional languages such
as Telugu or Hindi. The final output, including the transcript and summary, is presented through a
user-friendly interface, typically developed using Streamlit, allowing users to interact with the system
and obtain meaningful summaries with ease.

24
Text Preprocessing Module

This module cleans and prepares the raw transcript generated by the speech recognition system. The
goal is to make the text suitable for summarization by removing noise and standardizing its structure.
It begins with tokenization, which splits the transcript into words and sentences. Then, stopword
removal is performed to eliminate commonly used words (like "the", "is", "and") that do not
contribute much to the meaning. Lemmatization or stemming is applied next to reduce words to their
base or root form (e.g., "running" becomes "run"), which helps in reducing redundancy. Additionally,
special characters, extra spaces, and irrelevant symbols are removed. These steps ensure that the text is
clean, consistent, and semantically meaningful before summarization algorithms are applied.

Figure 4.3: Layered architecture of Video Trasncript Summarizer

Extractive Summarization

This module generates a summary by selecting the most important sentences from the transcript
without altering the original text. It uses the TextRank algorithm, which is a graph-based ranking
method inspired by PageRank. In this process, each sentence is represented as a node in a graph, and
edges are formed based on sentence similarity, typically measured using cosine similarity of TF-IDF
or sentence embeddings. The algorithm calculates importance scores for each sentence based on how
well it is connected to others. The top-ranked sentences are then selected to form the extractive
summary. This method is efficient, unsupervised, and works well for generating quick and factual
summaries.

25
 Abstractive Summarization Module

This module generates human-like summaries by rephrasing and condensing the transcript using deep
learning models. It utilizes transformer-based architectures such as T5 (Text-to-Text Transfer
Transformer) or BART (Bidirectional and Auto-Regressive Transformers). Unlike extractive
summarization, abstractive summarization understands the context of the input and generates entirely
new sentences to represent the core message. The model is fine-tuned on summarization datasets like
CNN/DailyMail and can produce fluent, grammatically correct, and semantically rich summaries. It
uses an encoder-decoder framework where the encoder processes the input text and the decoder
generates the output summary. This module is especially useful for long or complex videos where
extractive methods might fall short.

Figure 4.4: Implementation in Google Colab

4.5.1 Hardware and Software Requirements

o System : Pentium i3 Processor o Operating system : Windows 10/11, Ubuntu

o Hard Disk : 500 GB. 20.04+, macOS Monterey
o Monitor : 15’’ LED o Coding Language : Python
o Input Devices : Keyboard, Mouse o Tool : Jupyter Notebook, Google Colab
o Ram : 4 GB
o Graphics Card : NVDIA RTX 3060
26
 CHAPTER 5

TESTING

The purpose of testing is to discover errors. Testing is the process of trying to discover every
conceivable fault or weakness in a work product. It provides a means for users to determine, and
in case of failure, mitigate such risks. This process involves testing components, sub-assemblies,
assemblies and/or system out to ensure that it is functionally correct and reliable. Different
testing types of have been developed. Each test type is individually fulfilled a specific testing
requirement.

5.1 Testing Strategies

5.1.1 Unit Testing

Unit testing is a critical step in validating the reliability and correctness of individual components
(or "units") in your Video Transcript Summarizer project. Each function—from audio downloading to
text preprocessing and summarization—is tested in isolation to ensure it behaves as expected under
various conditions.

Test Strategy and Approach

The test strategy for the Video Transcript Summarizer project focuses on ensuring the accuracy,
reliability, and robustness of each individual module as well as the seamless integration of the overall
system. The approach begins with unit testing to validate core functionalities such as audio extraction,
speech-to-text transcription, text preprocessing, and both extractive and abstractive summarization
methods. Mocking and stubbing are used to simulate external dependencies like YouTube downloads
and API calls, allowing tests to run efficiently and independently. Following unit tests, integration
testing is conducted to verify the data flow between modules, ensuring that outputs from one
component correctly serve as inputs to the next. Special attention is given to edge cases, such as videos
with poor audio quality or multilingual transcripts, to assess system robustness. Additionally,
performance testing evaluates the responsiveness of transcription and summarization processes,
especially on different hardware setups. This layered testing approach guarantees that the system
performs accurately and consistently, providing reliable summaries across diverse video content.

Test Objectives

 Ensure that each module — including audio extraction, speech-to-text transcription,

preprocessing, and summarization — produces correct and expected outputs under various
input conditions.
 Confirm that all modules work together seamlessly, with proper data flow and error
handling between components, resulting in a coherent end-to-end summarization process.

27
 Assess the system’s ability to handle different video lengths, languages, and audio qualities

28
Features to be Tested

 Verify that the system correctly converts video audio into an accurate and coherent transcript
across different languages and audio qualities.
 Ensure the summarization module (both extractive and abstractive) generates clear, concise,
and relevant summaries that capture the main points of the transcript.
 Test the user interface for smooth handling of video URLs or uploads, proper error messages for
invalid inputs, and clear presentation of transcripts and summaries.

5.1.2 Integration Testing

Integration testing for the Video Transcript Summarizer project focuses on verifying the seamless
interaction and data flow between individual modules, ensuring they work together as a cohesive
system. This testing phase checks how well the audio extraction, speech-to-text transcription, text
preprocessing, summarization, and user interface components integrate and communicate with one
another. For instance, it validates that the transcript generated by the speech recognition module is
correctly passed to the preprocessing and summarization modules without loss or corruption of data. It
also ensures that user inputs from the interface trigger the entire pipeline smoothly, from downloading
videos to displaying summaries. Integration testing helps identify issues arising from module
dependencies, data format mismatches, or timing problems, thereby ensuring the system functions
reliably end-to-end in real-world scenarios.

5.1.3 Functional Testing

Functional testing for the Video Transcript Summarizer project involves verifying that each
feature performs according to the specified requirements and delivers the expected outputs. This
includes testing the core functionalities such as accepting a video URL or file input, successfully
extracting audio, and accurately transcribing speech into text. The preprocessing steps—like
tokenization, stopword removal, and lemmatization—are tested to ensure the transcript is cleaned and
formatted correctly. Both extractive and abstractive summarization modules are validated to confirm
they generate meaningful and coherent summaries that reflect the video content. Additionally,
functional testing covers the user interface to ensure users can easily input data, select summarization
options, and view or download results without errors. Error handling is also checked to confirm the
system provides appropriate messages for invalid inputs or processing failures. Overall, functional
testing guarantees that all features work as intended and provide a smooth user.

5.1.4 System Testing

System testing for the Video Transcript Summarizer project involves evaluating the entire
application as a complete and integrated system to ensure it meets all functional and non-functional
requirements. This testing phase validates the end-to-end workflow—from video input through audio
extraction, speech-to-text transcription, text preprocessing, summarization, and finally, output
presentation in the user interface. System testing checks the system’s performance under realistic
conditions, such as processing videos of varying lengths, multiple languages, and different audio
29
qualities. It also assesses usability, verifying that the interface is intuitive and responsive for users.
Additionally, system testing evaluates robustness by simulating errors like network interruptions,
unsupported video formats, or invalid URLs to ensure the system handles these gracefully. Security
aspects, such as preventing unauthorized access or injection attacks via input fields, may also be tested.
This comprehensive testing ensures the Video Transcript Summarizer operates reliably, efficiently, and
securely in real-world scenarios.

30
 5.1.5 White Box Testing

White box testing for the Video Transcript Summarizer project involves examining the internal
logic, code structure, and workflows of each module to ensure they function correctly and efficiently.
This testing approach requires a deep understanding of the underlying code, allowing testers to write
test cases that cover different paths, conditions, and loops within the preprocessing, transcription, and
summarization algorithms. For example, in the text preprocessing module, white box tests verify that
tokenization correctly splits sentences, stopwords are properly removed, and lemmatization accurately
normalizes words. In the summarization modules, tests ensure that the ranking algorithms and neural
model inference execute as intended without errors or inefficiencies. Additionally, white box testing
checks for proper exception handling, resource management, and integration points between functions.
By focusing on the internal workings, white box testing helps identify hidden bugs, optimize code
performance, and improve overall code quality, leading to a more reliable and maintainable
summarization system.

5.1.6 Black Box Testing

Black box testing for the Video Transcript Summarizer project focuses on validating the system’s
functionality without any knowledge of the internal code or implementation details. Testers interact
with the application solely through its inputs and outputs, such as providing video URLs or files and
examining the generated transcripts and summaries. This approach verifies whether the system meets
user requirements by checking if it correctly processes various types of videos, handles different
languages, and produces accurate and coherent summaries. It also tests the user interface for usability,
responsiveness, and proper error handling when given invalid inputs or unsupported formats. By
concentrating on the external behavior, black box testing ensures that the system delivers a seamless
and reliable user experience, effectively meeting functional specifications under diverse real-world
scenarios.

Test Results: All the test cases mentioned above passed successfully. No defects were
encountered during testing.

31
 CHAPTER 6

RESULTS AND DISCUSSION

The Video Transcript Summarizer project was evaluated using a variety of YouTube videos
across multiple domains such as educational content, interviews, and news segments. The system
successfully extracted audio using yt-dlp and transcribed speech to text using OpenAI’s Whisper
model with a high level of accuracy, even for videos with background noise or regional accents.
The transcription quality was particularly strong for English-language videos, with only minor
inaccuracies in the case of overlapping speakers or unclear pronunciations. For Telugu and Hindi
videos, Whisper handled multilingual transcription effectively, especially when combined with
translation modules for summarization in English.

Figure 6.1: Heatmap Visualization of Data

In the summarization phase, both extractive and abstractive methods were tested. The
extractive summarizer, using the TextRank algorithm, performed well by selecting key sentences
that preserved the original meaning, particularly for short to medium-length transcripts.
Meanwhile, the abstractive summarizer powered by the T5 transformer model generated concise,
grammatically correct summaries that paraphrased content rather than copying it directly. This
proved especially useful for long-form content, where users preferred quick insights over lengthy
transcripts. The combination of both techniques provided users with flexibility based on their

32
summarization needs.

Figure 6.2: Real vs. Predicted Text Comparison with Accuracy

33
Overall, the system demonstrated reliable performance across diverse test cases, with an average F1-
score of over 85% when comparing generated summaries to human-written references.

Figure 6.3: Average Accuracy Calculation Result

Execution time varied depending on video length and processing hardware, with GPU-enabled systems
offering significantly faster summarization. The project also showed strong usability, as the user
interface accepted various input formats and displayed results intuitively. These results indicate that
the Video Transcript Summarizer is a practical and efficient solution for generating high-quality
summaries from video content using NLP and machine learning techniques.

34
 CHAPTER 7

CONCLUSION

 The Video Transcript Summarizer project effectively demonstrates how natural language
processing and machine learning can be integrated to automate the summarization of spoken
content from video sources. By leveraging tools like yt-dlp for audio extraction, OpenAI’s
Whisper for speech-to-text transcription, and transformer-based models such as T5 for
abstractive summarization, the system provides users with a streamlined way to convert
lengthy video content into short, meaningful summaries. This is particularly useful for
students, researchers, and professionals who need to quickly grasp the essence of videos
without watching them in full.

 The project achieves its goal of supporting multilingual inputs, including regional languages
like Telugu and Hindi, thereby extending its usefulness in a diverse linguistic environment.
The inclusion of both extractive and abstractive summarization techniques offers flexibility
depending on user preferences—either highlighting key sentences or generating paraphrased
summaries. The system also handles a range of video content types effectively, from
educational lectures to informal conversations, proving its robustness across use cases.
Additionally, the user interface ensures accessibility and ease of use, even for non-technical
users.

 In conclusion, the Video Transcript Summarizer stands as a successful implementation of NLP

and machine learning techniques for real-world video analysis. It reduces the time and effort
required to consume content, enhances accessibility for non-native speakers, and offers a
scalable solution for summarizing large volumes of video data. Future improvements can focus
on enhancing summarization quality through domain-specific models, expanding support for
more languages, and integrating real-time summarization for live streams, making the system
even more powerful and versatile.

35
 REFERENCES

[11] J. Chavan, S. Palnitkar, A. Kaje, and C. Bhopnikar, “Transcript Summarizer of YouTube

Videos Using Deep Learning,” Journal of Artificial Intelligence Research & Advances, vol. 11,
no. 3, pp. 119–126, 2024.
[12] P. V. Kumari, M. C. Keshava, C. Narendra, P. Akanksha, and K. Sravani, “YouTube
Transcript Summarizer Using Flask and NLP,” Journal of Positive School Psychology, vol. 6,
no. 8, 2022.
[13] A. Anusuya, R. Monika, M. Maheswari, and R. M. S. Dr., “Multilingual YouTube Transcript
Summarizer,” International Journal of Advanced Research in Computer and Communication
Engineering, vol. 12, no. 4, 2023.
[14] P. Ilampiray, D. N. Raju, A. Thilagavathy, M. M. Tharik, S. M. Kishore, A. S. Nithin, and I.
Raj, “Video Transcript Summarizer,” E3S Web of Conferences, vol. 399, 2023.
[15] T. Lv, L. Cui, M. Vasilijevic, and F. Wei, “VT-SSum: A Benchmark Dataset for Video
Transcript Segmentation and Summarization,” arXiv preprint arXiv:2106.05606, 2021.

[16] M. Rochan, L. Ye, and Y. Wang, “Video Summarization Using Fully Convolutional
Sequence Networks,” arXiv preprint arXiv:1805.10538, 2018. [Online]. Available:
https://arxiv.org/abs/1805.10538
[17] Z. Ji, K. Xiong, Y. Pang, and X. Li, “Video Summarization with Attention-Based Encoder-
Decoder Networks,” arXiv preprint arXiv:1708.09545, 2017. [Online]. Available:
https://arxiv.org/abs/1708.09545
[18] M. Narasimhan, A. Rohrbach, and T. Darrell, “CLIP-It! Language-Guided Video
Summarization,” arXiv preprint arXiv:2107.00650, 2021. [Online]. Available:
https://arxiv.org/abs/2107.00650
[19] B. Zhao, X. Li, and X. Lu, “Hierarchical Recurrent Neural Network for Video
Summarization,” arXiv preprint arXiv:1904.12251, 2019. [Online]. Available:
https://arxiv.org/abs/1904.12251
[20] P. Solanki, “Automating Video Summarization (Conversational / Transcript
Summarization),” Medium, 2023. [Online]. Available:
https://medium.com/@prashants975/automating-video-summarization-conversational-transcript-
summarization-5e8afa67ea92
[21] B. Williams, “AI Summarize Video Transcript: Step-by-Step,” Insight7, 2023. [Online].
Available: https://insight7.io/ai-summarize-video-transcript-step-by-step/
[22] “Video Summarization Using OpenAI Whisper & Hugging Chat API,” Analytics Vidhya,
36
 2023. [Online]. Available: https://www.analyticsvidhya.com/blog/2023/09/video-
summarization-using-openai-whisper-and-hugging-chat-api/
[23] K. Porwal et al., “Video Transcription and Summarization Using NLP,” SSRN, 2022.
[Online]. Available: http://dx.doi.org/10.2139/ssrn.4157647
[24] P. Bedi, M. Bala, and K. Sharma, “Extractive Text Summarization for Biomedical
Transcripts Using Deep Dense LSTM–CNN Framework,” Expert Systems, vol. 40, no. 3, 2023.
[Online]. Available: https://doi.org/10.1111/exsy.13490
[25] “Automatic Summarization,” Wikipedia, 2023. [Online]. Available:
https://en.wikipedia.org/wiki/Automatic_summarization

37