Unity 2021.3 LTS+
Target editor version for all plugins.
Runtime audio, voice and AI
Build richer game experiences with offline tools for audio, voice and intelligent NPCs.
Denisitree is creating a toolkit for Unity™ developers who need runtime audio processing, speech recognition, local text-to-speech, and on-device AI without cloud services.
Built for Unity 2021.3 LTS+
C# API
Cross-platform
Made for the Unity workflow
C# API + Inspector
Use code or drag-and-drop components.
Local runtime workflows
Audio, voice and AI processing stay on-device.
Editor tools
Includes a Model Manager window for PocketMind.
Plugin suite
Built to work together in local, cross-platform Unity workflows.
Audio
SonicLoad
Runtime audio import & export
Import, decode, transcode and export audio at runtime for Unity projects that need flexible audio pipelines and raw buffer workflows.
Explore →
Speech to text
VoxScribe
Offline speech recognition
Capture live mic input or process AudioClip data into accurate offline transcripts with streaming results and language auto-detect.
Explore →
Text to speech
Vocalis
Offline text-to-speech
Create natural speech from game text without a network dependency, making NPC dialogue, accessibility and ambient audio easier to localize.
Explore →
Local LLM
PocketMind
On-device LLM inference
Use GGUF models for local reasoning, chat experiences and interactive NPC systems that run directly on the device.
Explore →Documentation
Guides and API reference for every plugin
Source-based guides for setup, usage patterns and current API boundaries.
SonicLoad
Runtime audio loading, WAV processing and playback.
Open docs →VoxScribe
Local speech recognition and microphone workflows.
Open docs →Vocalis
Local speech synthesis and AudioClip output.
Open docs →PocketMind
Local model generation, prompts and chat history.
Open docs →Ecosystem
One pipeline for offline audio, speech and agent logic.
Microphone
Files & buffers
Text
VoxScribe
PocketMind
Vocalis
AudioSource playback
Export via SonicLoad
The plugins are designed to work together as a local-first toolkit: ingest audio or speech, process it with speech recognition or LLM inference, then output speech or processed audio back into your Unity game systems.
Use cases
Designed for interactive and accessible Unity experiences.
Voice-controlled games
Build responsive gameplay around spoken commands and live in-game actions.
Smart NPCs
Compose private, local dialogue systems powered by on-device reasoning and speech output.
Accessibility
Support narration, captioning and speech-driven interaction without cloud dependencies.
In-game audio tools
Rig audio import and export pipelines for procedural content, editing and runtime generation.
Offline assistants
Build context-aware assistants for local, disconnected or privacy-sensitive experiences.
User-generated audio
Let players record, transcribe and process content without leaving the game session.
About
Tools that fit a local Unity workflow.
Denisitree is an independent project by Denis Treshchev focused on runtime audio, voice and AI tools for Unity developers. The tools are designed to work together across local audio, speech and on-device AI workflows.
Roadmap
Current focus is early prototyping and public interest gathering.
Concept
Market shaping and technical direction.
Core build
Runtime APIs and proof-of-concept experiments.
Private alpha
Limited validation with developers and early testers.
Public beta
Public access with documentation and sample scenes.
Release
Stable package shipment and full public availability.
FAQ
Questions before the first public release.
Windows, macOS, Linux, Android and iOS are planned for the runtime libraries, with Unity 2021.3 LTS+ as the target compatibility baseline.
Yes. The goal is local-first operation: VoxScribe works with offline models, Vocalis runs locally, and PocketMind is designed for on-device inference with GGUF models.
Licensing details will be published alongside the package terms. Feature scope and API behavior may change while the roadmap is validated.
There is no fixed date yet. The team is collecting public interest while the roadmap is being validated.
Performance depends on the device, model size and the task being run. Smaller models and simpler workflows are more practical for lower-end mobile hardware.
Yes. Reach Denis via Telegram or email with questions, feature requests and early-access interest.