Skip to main content
2026-04-18ChangelogPerformanceAI

Pure Lab 2.0 Performance Revolution: Millisecond Search, Externalized Backups & AI Asset Tagging

Pure Lab 2.0 brings major performance and architectural breakthroughs: a new FTS5 search engine for millisecond responses, externalized backup storage for enhanced safety, and deep Asset Event tagging with AI instrument detection.

Pure Lab 2.0 Performance Revolution: Millisecond Search, Externalized Backups & AI Asset Tagging

🚀 Dimensional Leap in Performance & Architecture

This month, Pure Lab 2.0 entered a deep optimization phase. We’ve not only solved performance bottlenecks for “massive libraries (500k+)” but also delivered multiple industrial-grade improvements in data safety and AI governance.

🔍 All-New Search Engine (Search V2)

For heavy collectors, we have completely refactored the search pipeline with a high-level FTS5 architecture.

  • Performance Miracle: Two-phase query optimization cuts keyword search time (with complex filters) from 8.6s to 339ms.
  • Advanced Audition Basket: The new search interface + audition basket lets you quickly filter, hold, and decide across massive material libraries.
  • BM25 Relevance Ranking: Industry-grade relevance algorithm ensures search results match your intent.

🧠 AI Audio Analysis Pipeline Rebuilt

To support more complex model combinations, we rewrote the on-device AI execution framework.

  • Embedding + Head Architecture: Converted Yamnet and AMC emotion recognition into “feature embedding + classification head” patterns, greatly improving inference reuse and speed.
  • AssetInstrument (Pro): Launched instrument recognition for professional material libraries—automatically tags tonal characteristics based on confidence scores.
  • GPU Deep Diagnostics: Integrated NVML tooling to monitor CUDA and TensorRT environments in real time, ensuring full hardware acceleration.

📂 Segment-Level Annotation (Asset Event)

Pure Lab’s governance boundary now extends from “file-level” to “segment-level”.

  • Asset Events: Annotate specific time segments within audio files. Combined with MaterialDetailPanel’s real-time waveform rendering, you can precisely manage every sample’s golden moment.
  • Multi-dimensional Integration: Custom Tags are now deeply integrated into search, extraction, and history rollback workflows, with global dictionary management.

🛡️ Externalized Backup & Stability

Data security is the lifeline of professional software.

  • Storage Migration UI: Move cover images and snapshots from your local library to a designated external directory, dramatically reducing main database size and corruption risk.
  • 500k-File Scale Scanning: Introduced jwalk parallel scanning and cancellable task stages—the system stays responsive even with 500k files.
  • State Persistence: Main window position, sidebar state, and search view settings are now globally persisted and remembered.

✅ This Month’s Update List

  • [x] Search V2 engine with FTS5 (millisecond response)
  • [x] Backup storage externalization migration tool & UI
  • [x] AI material instrument recognition (AssetInstrument)
  • [x] Asset Event tagging and visualization rendering
  • [x] Global custom tag dictionary sync
  • [x] 500k-scale library scan and write-back performance optimization