VoiceCast Builds Generative AI System for Kyodo News Broadcast Drafts
The design assumes the editor stays in the loop: AI drafts, humans verify, and the system flags possible hallucination by comparing output against the source article.
Reporting from 1 source: ASCII.jp.
VoiceCast Inc. has developed a system for Kyodo News that uses generative AI to turn newspaper articles into draft broadcast scripts. It also automates routine items such as heatstroke alert and heavy rain information. Staff fact-check, correct, and adjust the drafts. The system combines large language models with rule-based processing, and includes a check that flags content not present in the original article.
The system targets the part of broadcast writing that resists simple summarization. Newspaper text has to be reordered, reworded for reading aloud, and cut to fit a time slot before it can go on air, and that conversion has leaned on the experience of skilled staff. VoiceCast pairs large language models with rule-based processing to produce the draft, drawing on past editing examples. A separate function compares the generated script against the original article and raises an alert when it finds information that was not in the source.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.