TrueConf AI Secretary learns to separate speakers' voices in challenging acoustic conditions

Russian IT company TrueConf presented a new version of its corporate AI secretary that can separate the voices of video conference participants in final summaries even when working with mixed audio. The new technology allows creating such reports from any audio and video recordings, including those made in other communication solutions. Previously, TrueConf integrated smart features into PC and smartphone apps: active noise reduction, virtual backgrounds, speaker overlay, and face focus. In October 2024, the company introduced TrueConf AI Server, an on-premise AI secretary for mass transcription and meeting summaries that runs locally and does not transmit data to third-party servers. A major challenge in transcribing video conferences is that each participant needs a personal audio stream; if participants are in an open space or share a microphone, algorithms mix speech into a single block. The AI secretary now learns to separate voices and assign corresponding statements in the summary; users only need to name detected speakers. Demand comes from oil and gas sector clients, state companies, and executive committees holding large meetings. The development took about four months. The feature supports any audio and video files, including those from third-party systems, making TrueConf AI Server a universal transcription and summarization tool. Version 1.1 also adds full-text search across transcripts and customizable prompts for flexible summary preparation, such as key decisions, tasks, deadlines, and structured reports. TrueConf AI Server ensures transcripts are generated by local models within a closed customer circuit without access from company employees or administrators.