ActionPower, an artificial intelligence voice technology company, has introduced Korea's first feature that identifies meeting participants from a single registered voice sample and automatically records the speaker's name.
The company said on the 23rd that it had launched a voiceprint-based "Voice Matching" feature on Daglo, its AI transcription service. It is the first time in Korea that a service recognizing learned voices to automatically identify speakers has been offered to general users.
Once a user consents to the feature and replaces labels such as "Participant 1" in the meeting minutes with an actual name, the AI learns the characteristics of that voice, the company said. From then on, each new recording is compared against registered voices and the speaker's name is labeled automatically. Speaker identification accuracy measured in actual recording environments came to about 94%. Users can also collect and review a particular participant's remarks across multiple meetings, or generate a separate summary of that person's comments alone.
ActionPower is expanding its enterprise collaboration services on the strength of the speaker identification feature. It recently launched Team Plan, a service for teams, and obtained ISO 27001 certification, the international standard for information security management systems. The company plans to develop features that automatically organize each participant's assigned tasks and draft documents based on what was said in a meeting.
"What matters to office workers is not that a script remains, but that a record of 'who' said what remains," said Cho Hong-sik and Lee Ji-hwa, co-CEOs of ActionPower. "Through Voice Matching, we aim to grow into a one-of-a-kind service in which context accumulates naturally for each member of an organization, and to contribute to corporate AI transformation."