Resonant 0.5.0

Resonant 0.5.0

Resonant 0.5.0 turns local AI music generation and personal singing-voice conversion into first-class workstation capabilities.

Highlights

• Personal singing voices with Seed-VC: install and manage the optional local provider from Resonant, select an isolated source vocal and a clean 1–30 second reference, and add the converted performance directly to the Audio track.
• Consent and provenance by design: every conversion requires confirmation that the voice is owned or used with explicit permission. Resonant records source/reference hashes and conversion provenance beside the output.
• Visible provider controls: the main header now exposes labeled AI SONG and MY VOICE controls with install, ready, running, and error states instead of hiding providers behind icon-only actions.
• Managed local setup: Resonant detects existing managed ACE-Step and Seed-VC installations, offers installation when missing, and coordinates GPU use by stopping ACE-Step before Seed-VC conversion.
• Agent-accessible voice conversion: MCP clients can inspect Seed-VC availability and perform consent-gated conversions using workspace-confined WAV paths.

Seed-VC integration

• The managed Windows installation pins archived Seed-VC revision 51383efd921027683c89e5348211d93ff12ac2a8 and installs it separately under its GPL-3.0 license.
• Windows setup installs explicit PyTorch 2.4 CUDA 12.4 wheels and a minimal singing-inference dependency profile, avoiding malformed upstream requirement flags and unnecessary GUI/evaluation packages. Partial installations remain retryable and are not reported as complete.
• Installer progress strips terminal control codes, preserves a readable failure message, and removes temporary package caches after successful setup.
• The singing/F0 conversion path retains source melody, timing, phrasing, and pitch contour while transferring vocal identity from the reference recording.
• Source, reference, and output files are restricted to the configured workspace for MCP operations.
• An NVIDIA CUDA GPU with 6 GB VRAM is the practical minimum; 8 GB is recommended. Model weights download on first conversion and remain local afterward.

MCP additions

• get_voice_conversion_provider_status reports installation, runtime, hardware, and readiness state.
• start_voice_conversion, get_voice_conversion_status, collect_voice_conversion, and abandon_voice_conversion provide a durable job lifecycle that survives request timeouts and MCP reconnects. Stable idempotency keys prevent duplicate GPU work.
• convert_singing_voice remains available as a blocking compatibility tool.
• Conversion controls include diffusion quality, pitch shift, guidance strength, and 0.5×–2× performance-length adjustment.
• Capability metadata explains the voice-conversion workflow, hardware expectations, provider license, consent boundary, and isolated-vocal requirement.
• Newly generated Codex MCP configuration allows ten minutes per tool call so full-song local vocal analysis can complete on CPU; long AI generation and voice conversion still use the durable start/poll/collect workflows.

Validation

• Delivery-graph validation, TypeScript checks, and the complete unit/domain suite pass.
• Production renderer and MCP builds pass.
• MCP smoke validates all registered tools, including Seed-VC provider status and consent-gated conversion contracts.
• Electron smoke covers provider discovery, visible header states, the Seed-VC installation experience, responsive layouts, and the existing workstation workflows.

Downloads

• Resonant-Setup-0.5.0-Windows-x64.exe — standard Windows installer.
• Resonant-0.5.0-Windows-x64.exe — portable Windows build.
• SHA256SUMS.txt — checksums for verifying both executables.

The Windows executables are currently unsigned, so Windows SmartScreen may show a warning. Source installation instructions remain available in the README.