Hokkaido Cultural Broadcasting Streamlines Production of News Scripts and Videos Using Amazon Bedrock to Deliver More News at a Lower Cost

Established in 1972, Hokkaido Cultural Broadcasting serves its programs to the entire Hokkaido area as an affiliate of the Fuji Television Network. The company strives for technological innovation while promoting the charms of Hokkaido through community-based content. The station’s development initiatives on broadcasting and information processing are led by Ayuki Sugimoto in the Programming Department and Technical Lead concurrently at Hokkaido Cultural Broadcasting. “While I’m involved in data broadcasting and conventional programs, I also work on system development,” says Sugimoto. “Our company doesn’t have a dedicated system development department. Volunteers from the news reporting team do it.”

Previously, ratings, program content, and social media information were managed in separate formats, making data aggregation and analysis difficult. As a solution, Sugimoto created a business intelligence (BI) tool using Amazon QuickSight—which powers data-driven organizations with unified BI at hyperscale—to visualize, aggregate, and analyze the data at a low cost. “We didn’t know anything about AWS or the cloud to begin with,” says Sugimoto. “With support from AWS, we created a BI tool in just a couple of weeks. The support and plentiful hands-on materials from AWS made it easier for the beginner to get started.” 

Sugimoto’s transformative digital data visualization tool was met with great acclaim and won the Japan Commercial Broadcasters Association’s Technology Award in 2023. 

In the spring of 2023, Sugimoto started a new project involving news video production by the news department. “The news department wanted to publish more news online, but allocating production personnel was too expensive,” says Sugimoto. “So, we looked at ways to automate the process of creating news articles and videos.”

The news department writes articles and publishes them online by covering stories on press releases faxed by regional governments and companies daily. However, the team has a limited number of reporters, and the quality of their scripts and the time needed to create them depends on the skills of those reporters. Several more challenges prevented quick news distribution, including arranging announcers when creating audio and video based on the articles.

Sugimoto turned to AWS to streamline the production of news articles. Hokkaido Cultural Broadcasting had converted press releases to PDF format on a multifunction printer and emailed the PDF files to the news team. Sugimoto built a solution to perform optical character recognition on these PDF files and store the text data on Amazon DynamoDB, letting the team view the releases in text on a management screen. Members of the news team cover the story based on the releases and enter additional information through the management screen so that Amazon Bedrock—which is a fully managed service that offers a choice of high-performing foundation models from leading AI companies—can use all the data to output a news script. “We’ve configured the tool to output three model scripts for each news script,” says Sugimoto. “Our reporters choose the best script and edit it to complete the article. We use Anthropic’s Claude 2.1 as the foundation model, which delivers excellent accuracy and cost performance.”

With the station receiving over 1,000 press releases every month, the new solution simplifies how the team searches the releases. Team members can use keywords to search text information stored on Amazon DynamoDB with less effort.

Supplementing online news articles with videos adds value and opens other distribution channels, such as YouTube. Hokkaido Cultural Broadcasting therefore adopted Amazon Polly, which uses deep learning technologies to synthesize natural-sounding human speech, to narrate news videos. “When synthesizing speech, you must correct misreading and intonation,” says Sugimoto. “Because it was difficult for the operations team to add SSML tags to fine-tune the narrations, we simplified things for easier control. The team can perform operations like adjusting the time between sentences by adding line breaks from the management screen.”

The team also developed a tool to more easily create vertical videos, such as YouTube Shorts. The team members would previously perform this with a video editing application, but the time needed for the task could vary depending on each member’s proficiency. The station therefore built a user-friendly, browser-based solution that lets users choose a format, generate text and videos, and share them. This mechanism is designed using an AWS serverless environment and has passed internal security regulations.