Scholarlink

This study describes a simplified method of speaker diarization on Google Cloud Platform (GCP) by utilizing the PyAnnote toolkit. Using Cloud Functions, Vertex AI, and Kubeflow Pipelines, among other GCP services, we created an automated method to extract speaker information from video inputs. Based on Vertex AI, the system analyzes movies that are uploaded to Google Cloud Storage (GCS), extracts audio, and starts an inference pipeline. Vertex AI hyperparameter tuning shows how we can improve model performance even more and decrease the Diarization Error Rate (DER) using Vertex AI custom job. The system’s accuracy and efficiency are demonstrated by the experimental findings; going forward, user feedback methods will be the main emphasis for ongoing model improvement. This work highlights how machine learning techniques and GCP services may be effectively integrated to provide scalable speaker diarization solutions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scalable Speaker Diarization with PyAnnote: Deployment, Inferencing, and Optimization on Vertex AI

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Riyan Singh,
  • Sourabh Verma,
  • Om Prakash Verma,
  • Himanshu Gupta

摘要

This study describes a simplified method of speaker diarization on Google Cloud Platform (GCP) by utilizing the PyAnnote toolkit. Using Cloud Functions, Vertex AI, and Kubeflow Pipelines, among other GCP services, we created an automated method to extract speaker information from video inputs. Based on Vertex AI, the system analyzes movies that are uploaded to Google Cloud Storage (GCS), extracts audio, and starts an inference pipeline. Vertex AI hyperparameter tuning shows how we can improve model performance even more and decrease the Diarization Error Rate (DER) using Vertex AI custom job. The system’s accuracy and efficiency are demonstrated by the experimental findings; going forward, user feedback methods will be the main emphasis for ongoing model improvement. This work highlights how machine learning techniques and GCP services may be effectively integrated to provide scalable speaker diarization solutions.