Reasonate

AI audio production toolkit for REAPER, built on the ElevenLabs API. Replace voices in recorded dialogue, generate speech and multi-speaker conversations, dub material into other languages, fix single words in a recorded take, and design sound effects and music beds — all without leaving your DAW, all non-destructively.

Written in Lua (ReaScript + ReaImGui). No external runtime: REAPER's built-in Lua 5.4, curl, and your API keys.

Status: v1.0.0 — released via ReaPack and GitHub. Tested live on REAPER 7.71 / macOS arm64 (primary platform), REAPER 7.69 / Windows 11 and REAPER / Ubuntu Linux (aarch64). Community feedback on other setups is very welcome.

User guide: full illustrated documentation (all five modes, settings reference, costs, troubleshooting) at b451c.github.io/Reasonate/guide (source in docs/guide/). Questions and feedback: REAPER forum thread or GitHub Issues.

The five modes

Voice Replacement — swap the voice on recorded takes (speech-to-speech). Per-track voice mapping, async batch conversion with caching and rate-limit retry, multi-take variants, casts (role → voice presets), per-track recording with pre-roll, optional Voice Isolator pre-clean. Output lands on a separate "[AI]" track (folder child of the source); source items are never modified.

TTS — a small writing studio inside REAPER. Single-voice text-to-speech with the full ElevenLabs model lineup (v3 audio-tags palette with search, multilingual v2, turbo/flash), seeds, per-track defaults, voice presets — plus a multi-speaker dialogue sub-mode (text-to-dialogue endpoint): named cast, turn-by-turn lines, per-speaker voice settings, optional per-speaker track split via diarization.

Dubbing — translate and re-voice source material in another language while keeping the original timing. Pipeline: Scribe transcription (with speaker diarization and silence-aware chunking for long files) → LLM translation with editable context, glossary and style presets (bring your own key: Anthropic / OpenAI / Gemini / DeepSeek / xAI Grok / Mistral) → per-speaker voices (clone from source, similar voices, library pick, or voice design) → TTS with timing fit → forced-alignment splice onto per-speaker dub tracks. Per-segment regen, variants, inspector, cost tracking.

Repair — Descript-style word-level editing of recorded speech. Click words in the transcript, type the replacement, get a seamless splice: forced alignment for cut points, acoustic onset scanning, word-aligned loudness match, optional tempo match with élastique stretch. Replace / insert / delete.

SFX & Music — sound effects (/v1/sound-generation) and instrumental music beds (/v1/music) from a text prompt, or proposed by an LLM straight from your scene: select an item, the transcript is analyzed and you get editable, time-anchored candidates (effects plus an optional music bed) that insert exactly where the event happens. Results are grouped per generation with per-take preview, dedicated "SFX" and "Music" tracks, and an AI "New idea" rephrase per candidate.

Everything is non-destructive: source items are never altered (only metadata and colors), every insert/convert is wrapped in an undo block, and all generated audio is cached deterministically (repeating an identical request is free).

Screenshots

Voice Replacement - source untouched, the AI take lands on a linked child track:

Dubbing - a Spanish dub generated and fitted to the original timing:

Repair - click a word in the transcript, type the fix, apply:

SFX & Music - a complete sound package proposed from scene analysis:

More (all five modes, step by step) in the user guide.

Demo videos

• Dubbing demo (YouTube) - one podcast in, another language out, on the timeline
• Repair demo (YouTube) - fixing recorded words directly in the take

Both are also embedded in the user guide; direct mp4 downloads on the release page.

Requirements

• REAPER ≥ 7.0 (tested on 7.71, macOS arm64)
• ReaImGui ≥ 0.10.0.4 (via ReaPack)
• curl ≥ 7.76 (auto-detected; macOS/Linux system curl is fine)
• SWS extension - required for in-app audio preview (without it previews open in your system player)
• An ElevenLabs API key (paid tier recommended; Creator covers all features including dubbing and music)
• Optional: an LLM API key (Anthropic / OpenAI / Gemini / DeepSeek / xAI Grok / Mistral) for Dubbing translation, TTS Enhance and SFX scene analysis

Installation

ReaPack (recommended):

1. Install REAPER, ReaPack, ReaImGui ≥ 0.10 (Extensions → ReaPack → Browse packages → "ReaImGui") and the SWS extension.
2. Extensions → ReaPack → Import repositories… → paste: https://raw.githubusercontent.com/b451c/Reasonate/main/index.xml
3. Browse packages → search "Reasonate" → Install → Apply.
4. Run "Script: reasonate.lua" from the Action List → Settings → paste your ElevenLabs API key → Test → Save & fetch voices.

Manual install: copy the contents of scaffold/ to <REAPER resource path>/Scripts/Reasonate/, then Action List → "New action…" → "Load ReaScript…" → pick reasonate.lua. Optionally run _phase0_check.lua once to validate the environment.

Costs & privacy

Reasonate is a client for third-party AI APIs. Audio and text you process are sent to ElevenLabs (and, for translation/scene analysis, to the LLM provider you configure). API usage is billed to your accounts by those providers — character/credit costs are previewed in the UI where possible, and the deterministic cache makes repeated identical requests free. API keys are stored in REAPER's ExtState and passed to curl via a chmod 600 header file — never on a command line and never inside project files.

Repository structure

voice-changer-reaper/
├── LICENSE                  ← MIT (+ third-party notices)
├── README.md                ← this file
├── scaffold/                ← the plugin (copy this into REAPER Scripts)
│   ├── reasonate.lua        ← entry point
│   ├── modules/             ← core (api, cache, async, state, splicers, …)
│   │   ├── modes/           ← voice_replacement / tts / dubbing / repair / sfx
│   │   ├── gui/             ← ImGui views (30 files)
│   │   ├── llm/             ← anthropic / openai / gemini / deepseek / grok / mistral adapters
│   │   └── lib/json.lua     ← vendored rxi/json (MIT)
│   ├── workers/             ← curl wrappers, POSIX .sh + Windows .ps1 (async sentinel pattern)
│   ├── assets/fonts/        ← Inter Regular/SemiBold (SIL OFL 1.1, see OFL.txt)
│   ├── _phase0_check.lua    ← environment validator
│   └── reacast.lua          ← legacy entry-point shim (pre-rename installs)
├── scripts/                 ← quality gate: check.sh (syntax + luacheck +
│                              UI-English scan + headless tests) + git hook
├── tests/                   ← headless unit tests (lua5.4 tests/run.lua)
└── docs/guide/              ← the illustrated user guide (GitHub Pages)

Development

• Quality gate: sh scripts/check.sh — four gates (Lua syntax, luacheck with undefined-globals as errors, English-only UI string scan, headless unit tests). Install it as a pre-commit hook with sh scripts/install_hooks.sh.
• Tests run without REAPER: lua5.4 tests/run.lua (a minimal REAPER stub is vendored in tests/reaper_stub.lua).
• Code and all UI strings are English.
• Known limitations and troubleshooting live in the user guide.

License

MIT © 2026 Bartosz Sroczynski (falami.studio / b4s1c).

Bundled third-party components: Inter font (SIL OFL 1.1 — scaffold/assets/fonts/OFL.txt), rxi/json.lua (MIT). Reasonate is an independent project, not affiliated with or endorsed by Cockos (REAPER) or ElevenLabs.