Access AI
This product introduces AI-powered speech-to text and content discovery features for the YouTube Web Platform.
Pain Points
How can we increase YouTube's total hours of content watched by educational content viewers and ad revenue by using AI to enhance accessibility and usability through text-to speech features to exceed 70% accuracy of auto-generated captions and transcripts?
Problem Background
YouTube currently holds 52% of the video content market with 5.2 billion monthly users, competing with platforms like TikTok, Instagram, Facebook, and Netflix. While YouTube offers auto-generated captions and transcripts, their accuracy is only around 60-70%. Enhancing its speech-to-text AI could help YouTube stand out, particularly in the educational
In performing a detailed user survey of 13 participants, we cited two key pain points:
• Youtube’s Excessive forced Ad Experience significantly detracts from the viewing experience.
• Youtube’s existing Speech-to Text features of auto-generated captions and as-spoken transcripts have been cited to be incomplete, inaccurate or inconsistent transcriptions.
Supporting Data
Sometimes the auto-generated captions aren't accurate so I turn it off and make the volume higher to listen carefully
I watch a lot of international content and oftentimes there are no translations available or the one provided is not accurate
the transcript is not complete, it broke the sentences and does not separate the speech of speakers
content space, competing with platforms like Kajabi, Udemy, and Skillshare. Other platforms have integrated accessibility features like auto-captioning and multilingual subtitles.
Prioritized Features
Prioritized Features for Development
Based on the user persona analysis, prioritization metrics, and pain points identified, I have considered the following three use cases as the prioritized focus of the AI feature: 1
1. AI-powered Speech-to-text transcription feature for both content creators and viewers
2 Search suggestion based on history
3 Integrated speech-to-text translations of foreign languages
75% of people we spoke to mentioned they were unhappy with the Youtube captioning feature due to inaccurate, incomplete or inconsistent transcripts and captions.
Feedback
The preliminary user research analysis to validate this problem revealed that video captioning and transcripts affect a subset of users, educational content viewers and creators. This subset amounts for 86% of viewers within the United States and has become the second most popular genre on the platform.
Future Steps
• Continued training and development of AI model to integrate diverse dialects, accents and subject matter from niche or technical topics to improve the accuracy of AI speech-to-text features.
• Expand multi-lingual capabilities beyond Spanish and French to target South Asian demographic
Learnings
Product Manager Learnings:
RAHUL LOHANA
Designer Learnings:
Designer Learnings:
Jo Sturdivant
- Adapting to an Established Team: Joining the team in week 6 of 8 was challenging, as I had to quickly adapt to existing workflows, dynamics, and goals. This mirrors real-world situations where you often integrate into teams mid-project, and flexibility is essential.
- Work-Blocking for Efficiency: With only two weeks to complete the project, I learned the importance of a structured work-blocking system. This approach allowed me to manage my time effectively and meet deadlines under pressure.
- Making Data-Driven Design Decisions: Unlike my past projects, I had to rely on research conducted by others. This was a valuable experience in using pre-existing data to guide design decisions, helping me focus on the core insights without starting from scratch.
Developer Learnings:
Developer Learnings:
Vanady Beard
&
As the back-end developer, I learned how important it is to create efficient and reliable systems that support the entire application. This experience also taught me the importance of optimising the database and ensuring the backend is scalable and easy to maintain.
Developer Learnings:
Stephen Asiedu
&
As a back-end developer, I've come to understand the importance of being familiar with various database systems and modules. This knowledge enables me to build diverse applications and maintain versatility in my work. I've also learned that the responsibility for making the right choices rests on my shoulders, guided by my best judgement.
Developer Learnings:
&
Developer Learnings:
Maurquise Williams
&
- Process of Creating an MVP: Developing a Minimum Viable Product (MVP) taught me how to focus on delivering core functionalities balancing between essential features and avoiding scope creep.
- Collaboration in a Real-World Tech Setting: This experience taught me how to collaborate efficiently in a fast-paced tech environment, keeping the team aligned and productive, even while working remotely across time zones.
- Sharpening Critical Thinking and Problem-Solving Skills: This experience honed my ability to think critically and solve problems efficiently. By tackling challenges and finding quick solutions, I sharpened my decision-making and troubleshooting skills in a dynamic, real-world setting.
Developer Learnings:
Jeremiah Williams
&
All in all this experience was very awesome I learned that in coding with others being transparent is key
Developers Learnings:
Justin Farley
&
I learned how important communication is when working with a team. Communication provides understanding, advice, ideas, and much more. While working with the product team, I’ve found that communication keeps everything flowing smoothly. Working with a team also showed me that every member brings something different to the table and we all have to work together in order to align and meet our end goal.