Reusable voice presets
Create named zero-shot reference presets from WAV recordings you own or are permitted to use. Myna stores the managed reference, its SHA-256, settings, description, and preferred join mode.
Alpha / test build
Myna 0.9-alpha
Myna — Local Voice Studio for Long-Form Narration
Create reusable voice presets, break scripts into manageable chunks, generate multiple takes, approve the best results, adjust pauses, and assemble long-form narration locally on your Windows PC.
Developed by Amaleen Global LLC, the company behind CreatorKit Studio.
CPU-only FP32
Local generation
Windows 64-bit
Product overview
What is Myna?
Myna is a local voice cloning and narration application for creators who need fine-grained control over long-form voice production. It uses a reference recording to condition the Chatterbox Original English engine; it does not train a new model on that recording.
Instead of treating a complete script as one generation, Myna organizes narration into editable chunks. A chunk can hold multiple raw takes, so a creator can listen, regenerate only a weak section, approve a preferred take, tune the pause after it, and build a continuous preview without replacing the underlying recordings.
Current status: Myna 0.9-alpha is a public prerelease/test build, not a finished production or enterprise platform. Project formats, packaging, and compatibility may change before a stable release.
Built for long-form narration
Work section by section, not all-or-nothing
1
Add Voice Preset
2
Load Voice Model
3
Create Project
4
Auto Chunk
5
Generate Takes
6
Review & Approve
7
Adjust Pauses
8
Build Preview
9
Save / Export
Auto Chunk uses sentence-aware boundaries and preserves paragraph relationships. Individual chunks can be edited or split; adjacent chunks can be merged. Editing marks earlier takes as outdated rather than silently presenting them as current. Generation is serialized, and creators can generate missing takes in sequence or regenerate just the selected chunk.
Verified capabilities
A practical review-and-assembly workflow
Sentence-aware chunks
Turn a long script into manageable units, then edit, split, or merge where the narration needs a different rhythm. Paragraph-aware metadata remains available to the project.
Non-overwriting takes
Every generation becomes a new take. Regenerate with the same seed for a repeatable attempt or use a new seed for an alternative, while earlier raw WAV files remain available for review.
Approval and pause control
Choose the current approved take per chunk and adjust the following pause. Assembly uses those choices, so weak sections can be replaced without regenerating the full narration.
Matched Room Tone
The recommended join mode inserts low-level room tone between approved chunks to soften transitions. Assembly is reversible: it does not modify approved raw takes.
Optional light mastering
Light Documentary mastering is conservative, optional, and off by default. It creates a comparison output without changing raw takes and requires FFmpeg to be available.
Persistent projects
Saved project data includes the original script, chunk order and text, paragraph relationships, pause values, generated-take metadata, approved selections, seed history, voice settings, reference path, and model identity. Raw WAV files are stored alongside the project.
Provenance reports
Export a non-overwriting Markdown record with app and engine identity, reference hash, settings, take record, and post-production state. It supports record keeping; it is not proof of ownership or a commercial-use guarantee.
Local-first design
Your working files stay in the local Myna data area
Voice presets, managed reference recordings, projects, raw takes, assembled previews, exports, caches, and session logs are local installation data. After the required runtime and model components are installed, normal reference conditioning and speech generation use the local CPU and do not require a cloud inference service.
Myna is not fully offline from first launch: the public bootstrap needs internet access to download the application, isolated runtime, and pinned model payloads.
Share-build privacy boundary
The packaging process uses an explicit application allowlist and a fresh staging area. It is designed to reject personal audio, projects, preset identifiers, logs, caches, and development paths. Clean distributions start with no voice preset or reference audio.
- No personal presets
- No reference WAVs
- No projects or exports
- No personal logs or caches
Responsible voice use
Permission is part of the workflow
Use only voice recordings that you own or have permission to use. Myna is intended for legitimate narration, production, accessibility, experimentation, and creator workflows—not impersonation, deception, fraud, or unauthorized cloning. You are responsible for consent, applicable law, platform rules, and the licenses that govern your sources and outputs.
System requirements
Current Windows alpha test target
Minimum test target
- OS
- Windows 10 or 11, 64-bit
- CPU
- 64-bit, 4-core target; lower-end target not yet externally validated
- RAM
- 16 GB for the current alpha test target
- Load gate
- At least 6 GiB available physical memory immediately before model loading
- Storage
- Approximately 12 GiB free during installation
- GPU
- Not required; CPU-only FP32 inference
- Other
- Windows page file enabled; internet for initial component download
Recommended configuration
- CPU
- 6 cores or better
- RAM
- 24–32 GB
- Storage
- 15–20+ GiB free on an SSD
- Page file
- System-managed
- Mastering
- FFmpeg available for optional Light Documentary mode
CPU synthesis may be slower than real time. Generation is intentionally sequential, and performance and output quality vary with hardware, reference audio, text, and settings.
Installation
Small bootstrap, verified components
The prepared Windows bootstrap is a small native .NET Framework application. It is designed to download four required artifacts—the Myna application, an isolated Python 3.11 CPU runtime, and two pinned model archives—then verify every artifact with SHA-256 before extracting any of them.
Installation is per-user under %LOCALAPPDATA%\Amaleen Global LLC\Myna. Application files are separated from UserData; system Python is not installed or modified. A normal uninstall preserves UserData, while deleting it requires an explicit option.
The current alpha bootstrap is unsigned. Windows SmartScreen or antivirus reputation checks may show a warning. Myna does not bypass those warnings, browser confirmation, UAC, or execution-policy controls.
Engine & provenance
Pinned, attributed, and reviewable
Engine
Chatterbox Original English
Open-source engine developed by Resemble AI. Myna and Amaleen Global LLC do not claim ownership of Chatterbox.
Validated stack
chatterbox-tts 0.1.7
Python 3.11.9
PyTorch 2.6.0+cpu
torchaudio 2.6.0+cpu
Upstream license snapshot: MIT
Pinned revisions
Source
5de7a54aa4e5e2baadb0182dde554908b48b85c2Model
5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18Myna compares the installed engine environment with a frozen provenance snapshot before model use. A changed result prompts review without rewriting packages or blocking local work. The snapshot and exported reports aid reproducibility and record keeping; neither is legal advice nor a guarantee of commercial rights.
Current alpha limitations
What testers should expect
Variable results. Reference quality, script structure, and settings affect output. Speaker identity may not perfectly match the source.
Resource-heavy operation. Model loading uses substantial memory, and CPU generation can be slow.
Deliberately serial generation. Myna uses one model instance and one generation worker; running multiple copies is not recommended.
Pre-release packaging. The public installer is currently unsigned, Myna 0.9-alpha remains a test build, and compatibility may evolve before stable release.
Windows prerelease
Download Myna 0.9-alpha
Download the small bootstrap from the official Myna GitHub prerelease. It retrieves the application, isolated CPU runtime, and two model archives from GitHub Releases, then verifies their pinned SHA-256 values before installation. The multi-gigabyte payload is not hosted on CreatorKit Studio.
Download Myna BootstrapSHA-256: 5a970a8d5f8a4ab68a9549d68d5467a43e8a0687ee525e270e1ede52951da5f7
Windows alpha · about 3.8 GiB downloaded · about 4.7 GiB installed · 12 GiB free-space minimum · unsigned test build
Creator tool connections
Prepare the script before generation
FAQ
Questions about Myna
What is Myna?
Myna is a Windows application for local, reference-conditioned voice generation and long-form narration management. It combines voice presets, script chunks, multiple takes, approvals, pauses, assembly, and project records.
Does Myna run locally?
Yes. After setup, voice conditioning and generation run on the local CPU. Projects and audio stay in local Myna storage unless you choose to move or share them.
Does Myna need an internet connection?
Internet is required when running the public bootstrap to download the application, isolated runtime, and pinned model components. Normal generation does not require cloud inference after those components are installed.
Does Myna train a model on my voice?
No. The current workflow uses zero-shot reference conditioning: a permitted WAV reference guides a pre-trained speech model without training a new model for that voice.
Can Myna handle long scripts?
That is its primary workflow. Scripts are divided into sentence-aware chunks so you can generate, review, approve, and revise one manageable section at a time.
Can I regenerate only one sentence or chunk?
Yes. Select a chunk and generate another take using the same seed or a new seed. You do not need to regenerate the complete script.
Are old takes overwritten?
No. New generations create non-overwriting takes. If chunk text changes, prior takes can be retained but marked outdated so their status is clear.
Which operating systems are supported?
The current alpha targets 64-bit Windows 10 and Windows 11. macOS and Linux packages are not part of this build.
Do I need a dedicated GPU?
No. This validated stack uses CPU-only FP32 inference. A capable multi-core CPU, adequate available memory, a page file, and SSD storage are more relevant to this alpha.
Can I use any voice recording?
Only use a clean recording you own or have permission to use. You remain responsible for consent, laws, platform policies, and applicable licenses.
What speech engine does Myna use?
Myna 0.9-alpha uses Chatterbox Original English via chatterbox-tts 0.1.7, pinned to recorded source and model revisions. Chatterbox is developed by Resemble AI and distributed under an MIT license snapshot in this build.
Is Myna production-ready?
No. Myna 0.9-alpha is an alpha/test build. The installer is currently unsigned, generation speed and voice similarity can vary by hardware, reference audio, and script, and compatibility may continue to evolve before a stable release. A public prerelease is available for testing, but it should not yet be treated as a production-ready application.