How to automatically generate chapters from video and audio

Efficiently navigating long-form content like podcasts, webinars, and tutorials can be challenging without proper structure. Generating chapters solves this problem by dividing content into logical sections, enabling better accessibility and usability. Sieve's Transcript Analysis Function simplifies this process with two powerful features: Chapter Generation and Custom Chapters.

In this post, we’ll explore how these modes work, why they are valuable, and provide a quick guide to implementing them.

Why Generating Chapters Matters

Generating chapters for content provides several benefits for both creators and consumers:

Improved Navigation: Viewers can jump directly to sections of interest without manually scrubbing through the file.
Enhanced Accessibility: Structuring content into chapters makes it more digestible and reduces cognitive overload.
Increased Engagement: Clear segmentation encourages users to explore more of the content instead of abandoning it midway.
Professional Presentation: Adding chapters demonstrates a polished, user-centric approach to content creation.
SEO Benefits: Chapter markers can improve search rankings, particularly on platforms like YouTube.

By automating this process with Sieve’s Transcript Analysis Function, you can save time and effort while delivering high-quality results.

Overview of Sieve's Chapter Generation Features

Sieve's Transcript Analysis function offers two versatile modes for chapter creation:

1. Automated Chapters

This mode uses LLM-powered analysis to automatically segment your video or audio into logical chapters based on the transcript.

Ideal for:

Long-form content with natural breaks
Time-sensitive workflows
Quick, reliable chapter generation without manual input

2. Custom Chapters

This mode provides greater control by letting you manually define chapter titles or use an LLM to generate them.

Best for:

Projects requiring specific chapter names
Marketing videos with branded messaging
Content needing tailored chapter titles and timestamps

Available Modes:

"Strict" Mode: Generates only the chapters specified in your custom chapter list
"Extended" Mode (default): Combines your custom chapters with AI-generated ones for comprehensive coverage

Users can choose from various transcription backends like "groq-whisper" (default), "stable-ts," and "whisper-timestamped," as well as LLM backends including "gpt-4o-2024-08-06" (default), "gpt-4o-2024-05-13," and "gpt-4o-mini."

How to Generate Chapters Automatically?

Here’s a code snippet for automatically generating chapters from a video or audio file:

import sieve

file = sieve.File(url="https://storage.googleapis.com/sieve-prod-us-central1-public-file-upload-bucket/c4d968f5-f25a-412b-9102-5b6ab6dafcb4/52e169ed-9b30-4241-a956-591f2c6c30ab-chatgpt_tutorial_charlieChang.mp4")
transcription_backend = "groq-whisper"
llm_backend = "gpt-4o-2024-08-06"

transcript_analysis = sieve.function.get("sieve/transcript-analysis")
output = transcript_analysis.run(file,
	transcription_backend,llm_backend, prompt="",
	generate_summary=False, generate_title=False,
	generate_tags=False, generate_chapters=True)

output_list = list(output)
for chapter in output_list[2]['chapters']:
    print(f'{chapter['title']}:{chapter['timecode']} \n')

Results

A Dance Tutorial

Ellen and Brian | JENNIE 'Mantra' Dance Tutorial

File Duration- 11:34

Time taken to generate chapters- 13s

Chapters	Time-stamp
Introduction	00:00
Dance Tutorial Begins	00:18
Hand Movements	00:34
Adding Feet Movement	01:39
Combining Movements	02:05
Second Eight Count	03:06
Third Eight Count	05:39
Fourth Eight Count	08:01
Conclusion	11:11

Comment: The "Conclusion" chapter is not an appropriate title and is not time-synced correctly. All other generated chapters are excellent.

A Technical Tutorial

Charlie Chang | ChatGPT Tutorial: How to Use Chat GPT For Beginners 2024

File Duration- 27:05

Time taken to generate chapters- 24s

Chapters	Time stamp
Introduction	00:00
Account Setup	00:59
Limitations of ChatGPT	01:53
Basic Use of Prompts	02:32
Command Type 1: Facts	02:34
Command Type 2: Suggestions	03:51
Ad Break	04:58
Command Type 3: Language Translation	05:26
Command Type 4: Creative Writing	09:33
Command Type 5: Text Summarization	14:23
Command Type 6: List Generation	15:43
Command Type 7: Pros and Cons	16:47
Command Type 8: Studies and Quotes	17:29
Command Type 9: Feedback and Improvement	18:40
Command Type 10: Role Playing	19:55
Intro to ChatGPT & Demonstrations	21:37
Command Prompts for ChatGPT	22:53
Tips for Effective Use	25:54
Conclusion and Outro	27:03

Comment: Excellent!

A Podcast

Lex Fridman Podcast | Aravind Srinivas: Perplexity CEO on Future of AI, Search & the Internet

File Duration- 3:02:15

Time taken to generate chapters- 2m 10s

Chapters	Time-stamp
Exploring the Realms of Conversational AI: From Introduction and Concepts to Technical Insights and Functionality	00:00:00
Challenges and Strategies in AI, User Experience, and Product Design	00:04:12
Business Models & Market Dynamics	00:09:13
The Future of AI and Search Engines	00:13:13
User Experience Introduction and Considerations of Ad Breaks	00:29:36
Challenges in AI Output and Larry Page's Influence	00:34:03
Building a Product and Adapting to User Needs	00:38:22
Design Decisions	00:43:03
Influential Entrepreneurs: The Success of Bezos and Elon	00:46:52
The Impact of Open Source in AI and The Recipe for ChatGPT	00:56:13
Enhancing Post-Training Reasoning in Small Models with RLHF	01:11:07
Chain of Thought Reasoning	01:16:06
Future of AI Reasoning	01:20:14
The Intersection of Human Curiosity and AI: An Introduction to AI and Compute	01:24:20
AGI and Inference Limits	01:32:23
Ad Break	01:36:38
Perplexity and LLM Products	01:41:21
Challenges of Web Search Transformation	01:52:32
Exploring Truth and Prospects in AI Models and RAG	01:56:25
Improving Indexing and Snippets	02:01:15
Startup Challenges and Advice	02:21:25
Inspiration, Hard Work, and the Path to Education and Exploration	02:28:26
Ad Break, Search, and Knowledge Discovery: Navigating the Intersection of Marketing and Information Exploration	02:32:00
Perplexity and Curiosity	02:36:19
AI and Human Connection	02:50:43
Navigating the Intersection of AI and Human Values: Challenges, Biases, and Hope	02:55:42

Comment: Excellent! Employed min_chapter_length=210s parameter to control the number of chapters generated.

Real-World Use Cases

Podcast Hosts: Segment episodes into topics for better listener retention and easier sharing of specific segments.
Educators and Trainers: Enhance the learning experience by breaking down lectures into key concepts.
Marketers and Creators: Highlight promotional content or product demonstrations for greater audience engagement

Why Choose Sieve for Chapter Generation?

Efficient Automation: Generate accurate chapters instantly using advanced NLP and AI
Flexible Customization: Tailor chapters with manual or AI-assisted input
Multiple Backends: Choose from various transcription and LLM options
Simple Interface: Generate chapters easily, regardless of technical expertise

Conclusion

Sieve's chapter generation capabilities streamline content organization while enhancing accessibility and engagement. Start using the Transcript Analysis Function today to transform how your audience interacts with your content.

For support or questions, join our Discord community or contact us at contact@sievedata.com.