Turn-to-Diarize: Online Speaker Diarization Constrained by Transformer Transducer Speaker Turn Detection

TL;DR

This paper introduces Turn-to-Diarize, an efficient online speaker diarization system that leverages transformer transducer for turn detection and clustering, reducing computational costs and annotation efforts for streaming on-device applications.

Contribution

The system uniquely combines transformer transducer-based turn detection with constrained clustering, minimizing computational load and annotation requirements compared to traditional methods.

Findings

Reduces clustering computational cost due to sparse speaker turns.

Requires only speaker turn tokens during transcription, simplifying data annotation.

Suitable for real-time on-device speaker diarization.

Abstract

In this paper, we present a novel speaker diarization system for streaming on-device applications. In this system, we use a transformer transducer to detect the speaker turns, represent each speaker turn by a speaker embedding, then cluster these embeddings with constraints from the detected speaker turns. Compared with conventional clustering-based diarization systems, our system largely reduces the computational cost of clustering due to the sparsity of speaker turns. Unlike other supervised speaker diarization systems which require annotations of time-stamped speaker labels for training, our system only requires including speaker turn tokens during the transcribing process, which largely reduces the human efforts involved in data collection.