Home > Changelog

Changelog

Release history for Sound Forge Alchemy.

v4.6.0 — 2026-03-11

DJ Dual-Deck Instantaneous Playback and MIDI Fixes

JS.dispatch + JS.push dual-path architecture for sub-frame DJ playback response
AI cue detection engine (26+ cues per track) with AutoCueWorker Oban job
Stem loop decks for per-stem loop control during live mixing
Crossfader curve modes: linear, constant power, sharp cut, slow fade
SMPTE/bar-beat transport display at 60fps
Master sync with phase-aligned pitch-lock between decks
Built-in metronome with headphone cue routing
Chef AI set builder for harmonic/energy-optimized track sequencing
Virtual controller (software MIDI surface in-browser)
Chromatic pads with auto-cue integration and velocity sensitivity
Controller preset import/export as JSON

MIDI Fixes:

Fixed ETS key collision in DeviceManager – composite port_id ("input:N" / "output:N") replaces raw num
Added Dispatcher to supervision tree (was missing from application.ex)
Wrapped phx-change select elements in <form> tags (LiveView requirement)
Fixed resolve_user_id guard (changed from is_binary to is_integer)
Fixed Mapping schema PK mismatch (removed UUID override, uses integer serial)

WebSocket Optimization:

Fixed 30Hz WebSocket flood – debug_log handler guarded on debug_panel_open
BPM throttle handler with 5-second interval (local JS.dispatch + throttled server sync)

Bug Fixes:

Fixed ArgumentError in toggle_play / set_hot_cue / trigger_cue – to_string type normalization at handler boundary

Database:

New midi_results table
New chord_results table
New auto_midi_chord user settings columns

v4.5.0 — 2026-02-26

Audio-to-MIDI, Chord Detection, Piano Roll, Audio Warping

Audio-to-MIDI conversion via basic-pitch
Chord detection via librosa chroma analysis
Piano roll visualization component
Audio warping via pyrubberband
Auto-pipeline extensions for MIDI and chord stages
16 competitive response stories
45 files changed, +4,731 lines

v4.4.0 — 2026-02-25

Prototype Sandbox, DevTools Panel, Platform Admin Library

/prototype sandbox (Components, DevTools, UAT, LLM tabs)
DevTools floating panel component
UAT fixture/scenario helpers (runtime env guard)
CombinedLibraryLive at /platform/library (platform_admin)
Accounts context (list_users, update_user_role)
Tracks context (list_all_tracks, paginated search)
41-page GitHub Pages documentation site (Jekyll, just-the-docs dark theme)

v4.3.0 — 2026-02-25

Multi-LLM Agentic System

LLM adapter/routing layer with 6 specialist agents
Agent framework with Orchestrator dispatch
Chat UI for agent interaction
Health worker for provider monitoring
707 tests passing

v4.1.0 — 2026-02-25

Azure Container Apps Production Deployment

Azure Container Apps deployment with SSL termination
Azure Container Registry image build pipeline
DAW editor fixes and stability improvements
Comprehensive GitHub Pages documentation
SoundForge.Release.migrate/0 for production migration runner
Docker multi-stage build optimization

Bug Fixes:

Fixed DAW component routing (LiveComponent instead of standalone LiveView)
Fixed stem file_path stored as relative (not absolute) for clean /files/stems/... URLs
Fixed pipeline_complete?/1 to check only triggered stages

v4.0.0 — 2026-02

lalal.ai Full Cloud Stem Separation Integration (82 files, +12,398 lines)

Cloud stem separation via lalal.ai API (9+ stem types)
60-second preview before full processing
lalal.ai quota management dashboard
GET /api/lalalai/quota endpoint
POST /api/lalalai/cancel and cancel-all endpoints
Per-user lalal.ai API key storage (encrypted)
Engine toggle: local Demucs vs. cloud lalal.ai
LalalaiWorker Oban background worker
Voice pack service for voice transformation
Voice change and voice clean workers

Admin:

Admin dashboard with SaaS role hierarchy (user, admin, platform_admin)
Audit logging for role changes, deactivations, config changes
User management UI

v3.0.0 — 2026-01

Analysis Expansion (24 files, +4,712 lines)

5 D3.js visualization hooks: AnalysisRadar, AnalysisChroma, AnalysisBeats, AnalysisMFCC, AnalysisSpectral
MFCC (13 coefficients) and chroma (12 pitch classes) feature extraction
Spectral centroid, rolloff, bandwidth, contrast features
Extended AnalysisResult.features JSONB column for high-dimensional data
Beat frame detection and beat grid timeline visualization
Key detection confidence score
Analysis export as JSON (GET /export/analysis/:track_id)

v2.0.0 — 2025-12

Multi-LLM Agent System

Six specialist agents: TrackAnalysis, MixPlanning, StemIntelligence, CuePoint, Mastering, Library
SoundForge.Agents.Orchestrator with keyword-based auto-routing and capability-based dispatch
SoundForge.LLM.ModelRegistry GenServer with ETS storage and 5-minute health checks
Per-user LLM provider configuration with encrypted API keys (Cloak.Ecto AES-256-GCM)
Support for: Anthropic, OpenAI, Google Gemini, Ollama, Azure OpenAI
System-level LLM provider fallbacks from environment variables
DAW LiveComponent with multi-track editor and MIDI export
DJ LiveComponent with two-deck mixer, BPM sync, and loop controls
MIDI hardware mapping (Pioneer DDJ-200, Traktor Kontrol S2 presets)
OSC server for DAW integration

v1.0.0 — 2025-11

Initial Phoenix/Elixir Port

Full port from Node.js/TypeScript to Elixir/Phoenix 1.8
Phoenix LiveView dashboard replacing React + Socket.IO
Oban background jobs replacing Redis + BullMQ
Erlang Port architecture for Demucs and librosa (replacing Node child_process)
4-stem Demucs separation (htdemucs, htdemucs_ft, mdx_extra)
Basic audio analysis: tempo, key, energy, spectral centroid
Spotify metadata fetch + spotdl audio download
Phoenix scope-based authentication (phx.gen.auth)
PostgreSQL schema: tracks, download_jobs, processing_jobs, analysis_jobs, stems, analysis_results
653 ExUnit tests, 0 failures