PLAY, inc. provides "KRONOS DRIVE," a storage service specializing in the management and distribution of media data.
This service is used by broadcasting stations and those who operate video streaming services.
With the speed at which video content is consumed increasing rapidly these days, we spoke to Kenichi Maruyama, Tech Lead at PLAY CLOUD Headquarters, about how voice recognition is being used to accelerate content production and distribution.

Issues/background

In the video streaming industry, where consumption speed is fast
Leveraging voice recognition to support rapid delivery

Our company operates many of Japan's largest video distribution services, and in recent years, the speed at which content is consumed has increased more than ever before, with viewing styles such as "double-speed viewing" becoming popular, especially among the younger generation.
Therefore, in order to accelerate content production and distribution, rather than simply distributing already completed content, we have begun offering KRONOS DRIVE as a media supply chain platform to support content holders such as broadcasters and customers who provide video streaming services.

KRONOS DRIVE is a storage service similar to Google Drive that specializes in managing and distributing media data. Its main uses include managing data such as videos and images, sharing them externally, simple editing and format conversion, and submitting them to distribution services, all of which are necessary for video distribution.

Previously, KRONOS DRIVE was able to convert existing subtitle data, but there were many cases where it was necessary to create subtitles from scratch, and so we decided to use voice recognition technology to make it faster and easier to create subtitles in such cases.

Key Drivers of the Adoption

High recognition accuracy and "Charging only for the spoken part" are attractive features

We chose the AmiVoice API because it has been widely adopted in the broadcasting industry and has very high recognition accuracy.
Another feature not found in other services is that only sections that are recognized as actual speech are charged (silent sections are not charged), which we believe is a major advantage in that it allows us to optimize costs.
Also, the API interface was simple, so I didn't find it particularly difficult to implement.

Introduction Results

Subtitling made easy and fast

I feel that there are strengths and weaknesses depending on the content, but since it is specialized in Japanese voice recognition, I feel that it generally has high recognition accuracy.
Generative AI services can sometimes result in translation errors, but the AmiVoice API's faithful transcription of the original audio suited our needs.
Also, from a developer's perspective, I think it was great that the implementation was simple, as even long content of several hours can be processed with a single request without having to be split up.

Although this feature has only just been released, it is already being used by several customers.
On the other hand, it is difficult to distribute subtitles transcribed using a speech recognition API as is, and human review and correction is inevitably required.
KRONOS DRIVE offers a variety of features to make these tasks easier.

 

Future prospects

We would like to expand support for KRONOS DRIVE's subtitle transcription feature to languages ​​other than Japanese.
The AmiVoice API currently supports English, Chinese, and Korean, but we hope to support more languages ​​in the future.
We also receive requests from customers to transcribe audio that contains a mixture of multiple languages, such as Japanese and English, so we would be happy if you could accommodate this request.