github ventura8/Whisper-Pro-ASR v1.0.3
v1.0.3 — Dynamic Language Identification

latest releases: v1.5.1, v1.5.0, v1.4.3...
6 months ago

This update introduces a major overhaul of the language detection system, replacing the multi-zone voting mechanism with a more robust and efficient dynamic chunking approach designed specifically for long-form media like movies.

🚀 Key Improvements

🎯 Dynamic Language Detection

  • Automated Chunk Scaling: The identification engine now calculates an optimal sample size based on the total media duration.
  • Minimum 5-Minute Sample: Ensures that even short clips have enough context for high-confidence identification.
  • Movie Optimization: For a standard 4-hour movie, the system now extracts a 12-minute representative sample (5% of duration), significantly improving accuracy over 30-second snippets.
  • Single-Pass Efficiency: Replaced the iterative 15-scan voting system with a single, high-fidelity inference pass. This reduces FFmpeg overhead and simplifies the processing pipeline.

⚡ Performance & Stability

  • Reduced Latency: By eliminating multiple parallel extraction tasks and redundant inference cycles, /detect-language responses are now significantly faster.
  • Lower Memory Pressure: The single-pass approach ensures more predictable VRAM/RAM utilization during the identification phase.
  • FFmpeg 8.1.0 Refinement: Optimized the extraction command to use direct seek and duration flags for near-instantaneous chunk creation from any media container.

🛠️ Internal Changes

  • Version: Bumped to 1.0.3.
  • Logic: Refactored modules/language_detection.py to remove the multi-zone voting engine.
  • Tests: Updated unit tests to verify dynamic scaling calculations across various file lengths.

📦 Deployment

Update your docker-compose.yml to use the 1.0.3 tag (or latest) and restart the service:

docker compose pull
docker compose up -d

Optimized for high-precision identification in complex media libraries.

Don't miss a new Whisper-Pro-ASR release

NewReleases is sending notifications on new releases.