Problem Definition

In the existing LCK (League of Legends Champions Korea) broadcasts, the process of manually selecting highlight moments was cumbersome and required a significant amount of labor. Therefore, there was a desire to develop a program that automatically extracts highlights from videos while reducing the need for human intervention.

Approach

• Video Team
• Chat Team
• Audio Team
• Specification Team

Through the combined efforts of these various teams, the task of extracting highlights from videos and saving them was carried out automatically.

Role of the Audio Team

Role: Leader

Team Size: 3 members

Initial Phase

Initially, the team planned to use STT (Speech To Text) to convert audio files into text and extract the names of specific players or game characters for frequency measurement. Additionally, the team aimed to measure the pitch of the audio and combine all these factors to assign a final score. STT services such as Google's STT and Naver CLOVA's API were considered for this phase.

However, this plan faced several limitations and was deemed unfeasible. The reasons include:

1. Cost Issues: The majority of video sources were over 2 hours in duration, making it impossible to translate them all using free versions of STT programs. Due to budget constraints, this approach was not viable.
2. Technical Challenges: The audio files provided by the Specification Team contained all sounds (commentator voices, in-game sounds, audience noise, etc.) in a single file. Separating these sounds proved to be extremely difficult, making STT impractical.
3. Game Terminology: "League of Legends" features specific in-game terminology and proper nouns that STT services were unfamiliar with, resulting in numerous translation errors.

Therefore, STT usage was ultimately ruled out.