Skip to content

🇪🇸 Español: Consulta el historial de cambios en español en .github/locales/es/CHANGELOG.md.

title: Changelog description: Change log for the Voice2Machine project. ai_context: "Versions, Change History, SemVer" depends_on: [] status: stable

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • Zero-Copy Audio Engine: New ZeroCopyAudioRecorder in Rust using shared memory (/dev/shm) for true zero-capacity transfers.
  • Hallucination Detection: Heuristic filters and quality parameters (no_speech, compression_ratio) in StreamingTranscriber to reduce erroneous Whisper outputs.
  • Performance Metrics: Inference latency tracking in logs for detailed diagnostics.

Changed

  • Advanced Whisper Config: Increased beam_size and best_of to 5 for higher transcription quality in the "large-v3-turbo" model.
  • VAD Optimization: Adjusted default threshold to 0.35 to reduce false positives from ambient noise and breathing.
  • Memory Management: Forced CUDA cache reset (torch.cuda.empty_cache()) when unloading models to effectively free VRAM.
  • Code Hygiene: Import refactoring and linting error fixes (ruff) throughout the backend codebase.

Planned

  • Support for multiple simultaneous transcription languages
  • Web dashboard for real-time monitoring
  • Integration with more LLM providers

[0.3.0] - 2026-01-23

Added

  • Feature-Based Architecture: Total restructuring into self-contained modules in features/ (audio, llm, transcription).
  • Orchestration via Workflows: Introduction of RecordingWorkflow and LLMWorkflow to decouple business logic from the monolithic legacy Orchestrator.
  • Strict Protocols: Implementation of typing.Protocol for all internal services, allowing easy swapping of providers.
  • Modular API: Package structure in api/ with separate routes and schemas.

Changed

  • Elimination of Orchestrator: services/orchestrator.py has been decomposed and removed.
  • Infrastructure Refactoring: The infrastructure/ folder has been integrated into each corresponding feature.
  • Core and Domain: Simplified and moved to shared/ and local interfaces.

Removed

  • Legacy Audio Tests: Removal of obsolete tests for the Rust extension.
  • System Monitor: System telemetry removed for core simplification.

[0.2.0] - 2025-01-20

Added

  • FastAPI REST API: New HTTP API replacing the Unix Sockets-based IPC system
  • WebSocket streaming: /ws/events endpoint for real-time provisional transcription
  • Swagger documentation: Interactive UI at /docs for testing endpoints
  • Orchestrator pattern: New coordination pattern that simplifies workflow
  • Rust audio engine: Native v2m_engine extension for low-latency audio capture
  • MkDocs documentation system: Structured documentation with Material theme

Changed

  • Simplified architecture: From CQRS/CommandBus to more direct Orchestrator pattern
  • Communication: From binary Unix Domain Sockets to standard HTTP REST
  • State model: Centralized management in DaemonState with lazy initialization
  • Updated README.md with new architecture

Removed

  • daemon.py: Replaced by api.py (FastAPI)
  • client.py: No longer needed, use curl or any HTTP client
  • Binary IPC protocol: Replaced by standard JSON

Fixed

  • Startup latency: Server starts in ~100ms, model loads in background
  • Memory leaks in WebSocket connections

[0.1.0] - 2024-03-20

Added

  • Initial Voice2Machine system version
  • Local transcription support with Whisper (faster-whisper)
  • Basic LLM integration (Ollama/Gemini)
  • Unix Domain Sockets-based IPC system
  • Hexagonal architecture with ports and adapters
  • TOML-based configuration