Create README.md
Browse files
README.md
CHANGED
@@ -1,13 +1,75 @@
|
|
1 |
---
|
2 |
-
title: Project2
|
3 |
-
emoji: π»
|
4 |
-
colorFrom: gray
|
5 |
-
colorTo: blue
|
6 |
-
sdk: gradio
|
7 |
-
sdk_version: 5.18.0
|
8 |
-
app_file: app.py
|
9 |
-
pinned: false
|
10 |
license: apache-2.0
|
|
|
|
|
11 |
---
|
12 |
-
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: apache-2.0
|
3 |
+
title: 'HeadlinesGen:Partmone '
|
4 |
+
sdk: gradio
|
5 |
---
|
6 |
+
Final Project: Part One
|
7 |
+
Headlines Generation Project:
|
8 |
+
This project is all about making it easier to come up with the perfect headline. The goal is to
|
9 |
+
create an app that helps generate clear, engaging, and relevant headlines for articles in both
|
10 |
+
Arabic and English. Instead of struggling to summarize an article, users can get an instant, wellmatched headline with just a click. The app is designed to be simple and efficient, allowing
|
11 |
+
users to choose their preferred language and quickly get the best possible headline with
|
12 |
+
generated audio without the hassle.
|
13 |
+
Pipelines:
|
14 |
+
Headlines Generator:
|
15 |
+
β’ The app generates headlines based on the provided article text in English using the
|
16 |
+
Michau/t5-base-en-generate-headline model.
|
17 |
+
β’ If the input is in English, the headline is generated directly in English.
|
18 |
+
Translation:
|
19 |
+
β’ If the input is in Arabic, it is first translated to English using Helsinki-NLP/opus-mt-aren, then the headline is generated in English, and finally, it is translated back to Arabic
|
20 |
+
using Helsinki-NLP/opus-mt-en-ar.
|
21 |
+
β’ This ensures the app provides a headline in the desired language, whether the input
|
22 |
+
is in English or Arabic.
|
23 |
+
Text-to-Speech:
|
24 |
+
β’ The app supports text-to-speech conversion for both English and Arabic.
|
25 |
+
β’ Arabic: To convert the Arabic headline to speech using facebook/mms-tts-ara.
|
26 |
+
β’ English: To convert the English headline to speech using microsoft/speecht5_tts.
|
27 |
+
How to use the interface:
|
28 |
+
The app uses Gradio for building a simple interface where users can select the language, input
|
29 |
+
text, and receive the generated headline with an audio.
|
30 |
+
Select Language: Choose between Arabic or English.
|
31 |
+
Enter Article Text: Type or paste the article text from which you want to generate a
|
32 |
+
headline.
|
33 |
+
Get Headline: Once the button is clicked, a headline will be generated.
|
34 |
+
Get Audio: An audio of the headline also will be generated.
|
35 |
+
Example Input:
|
36 |
+
Select Language: English
|
37 |
+
Text/Article: " Greenhouse gas emissions, primarily carbon dioxide (CO2) and methane
|
38 |
+
(CH4), are the main drivers of global climate change. Human activities, such as burning fossil
|
39 |
+
fuels for energy, deforestation, and industrial processes, have significantly increased the
|
40 |
+
concentration of these gases in the atmosphere. According to the Intergovernmental Panel
|
41 |
+
on Climate Change (IPCC), CO2 levels have risen by over 50% since the pre-industrial era,
|
42 |
+
contributing to rising global temperatures. "
|
43 |
+
Expected Output (A headline in English): Global Warming - The Main Driver of Global
|
44 |
+
Climate Change.
|
45 |
+
An audio that read the headline.
|
46 |
+
Why We Chose These Models:
|
47 |
+
Headlines Generator:
|
48 |
+
We selected Michau/t5-base-en-generate-headline for headline generation because it has
|
49 |
+
been trained on a collection of 500,000 articles with corresponding headings, making it wellsuited for this task. The model is specifically designed to generate concise and relevant
|
50 |
+
headlines from article text.
|
51 |
+
Translation:
|
52 |
+
We use the Helsinki-NLP/opus-mt-ar-en and Helsinki-NLP/opus-mt-en-ar models to
|
53 |
+
translate back and forth between Arabic and English. Since the main headline generation
|
54 |
+
model does not support Arabic, we had to find a way to generate headlines in Arabic.
|
55 |
+
To achieve this, we first translate Arabic input into English so that the headline generation
|
56 |
+
model can process it and produce a proper English headline. Since we want the headline to
|
57 |
+
match the article's original language, we then translate the generated English headline back
|
58 |
+
into Arabic before outputting it.
|
59 |
+
Because translation accuracy is crucial for maintaining the meaning of the headlines, we
|
60 |
+
carefully selected our translation models based on popularity and high download rates,
|
61 |
+
ensuring they are optimized specifically for Arabic and English.
|
62 |
+
Text-to-speech:
|
63 |
+
Arabic Text-to-Speech: We selected facebook/mms-tts-ara for Arabic text-to-speech after
|
64 |
+
evaluating multiple models. This model provided the best pronunciation and clarity, ensuring
|
65 |
+
high-quality Arabic speech.
|
66 |
+
English Text-to-Speech: We chose microsoft/speecht5_tts for English text-to-speech due to
|
67 |
+
its superior pronunciation and clarity. As a Microsoft-developed model, it leverages state-ofthe-art technology to produce natural and highly reliable speech, making it an ideal choice
|
68 |
+
for our project.
|
69 |
+
Special Measures Taken to Support the Arabic Language:
|
70 |
+
Since our chosen headline generation model Michau/t5-base-en-generate-headline only
|
71 |
+
supports English, we implemented a translation pipeline to ensure that Arabic inputs can be
|
72 |
+
processed effectively.
|
73 |
+
To achieve this, we use the Helsinki-NLP/opus-mt-ar-en model to translate Arabic text into
|
74 |
+
English before generating a headline. Once the headline is created in English, we then use
|
75 |
+
the Helsinki-NLP/opus-mt-en-ar model to translate it back into Arabic
|