YOUR LOCAL AI STUDIO

Make something yours.

Text to speech · Speech to text.

On your device

Voice studio

VoxCPM2

Text to read

Khmer · English · MultilingualLoad model to count tokens

Try

Default voice

VoiceManon Voice

Choose speaker audioWAV, MP3, OGG · 3–20 seconds · saved only on this device

Saved as your default voice on this device.

I own this recording or have permission to use this voice.

WAV48 kHz audio

Load the model to get started.

Voice model

VoxCPM2 · runs on this device

Compute deviceChecking your browser…

Checking your browser…

First download ≈ 1.9 GB. Kept in this browser when space allows.

Fine-tune

Voice strength 2.0

NaturalCloser to text

Quality 10

FasterMore detail

More controls +

Seed Blank = 42

Play when ready

Your text and audio stay on this device.

Use only voices and recordings you own or are licensed to use. Get permission before cloning a real person.

VoxCPM2 supports Khmer, but its publisher does not disclose training-data sources or consent details. Training-data details.

Your creations

Your next voice starts above.

Listen, save, and make something new.

Page 1 of 1

Audio to words.

STT

Choose audioWAV, MP3, M4A, WebM

Or record a short clip.

Choose audio to begin.

Your transcript

↓ TXT ↓ SRT

Transcription

Model

English and Chinese speech. Small models can miss words; review the result.

Audio language

Google model file

Choose Gemma 3n E2B or E4B with “Web” in the filename. Audio up to 30 seconds.

Get Google model ↗

Google requires its model terms. Import the downloaded file; it stays here.

Checking saved files…

Model files stay in browser storage, not your Downloads folder.

Your files stay on this device.

ChatCraft

What can you help me with?

For fun: jokes, short stories, and creative ideas.

Start a new chat?

This clears the visible messages. ChatCraft never sends earlier messages to the model.

Your message

Type a message to see its token count.

A private conversation, on this device.

ChatCraft model

Model

For fun and casual use only. Answers may be inaccurate.

Google model file

Import a web-compatible Gemma 3 or 3n model.

Get Google model ↗

Google requires its model terms. Import the downloaded file; it stays here.

Checking saved files…

Each reply is fresh. Earlier messages are not sent to ChatCraft, and chats are not saved.

No account. No cloud inference.

Settings

Language, appearance, Wi-Fi and local storage.

Language

Appearance

Device & Wi-FiShared API · Start here for Wi-Fi & offloadOff

One connection for Voice, Transcribe and Chat.

Frees memory. Keeps downloaded files.

Host Wi-Fi API

The REST/JSON API runs on this computer. Native apps and command-line clients can call it; browser pages cannot host it or call it directly. Download and extract the server, then run the command below.

Download local API servernode companion.mjsOpen local studio ↗

Network address

Off

API addressAccess token

Keep this tab open. Anyone with this token on your Wi-Fi can send requests. Use a trusted network; HTTP is not encrypted.

Wi-Fi mode shares API results with authorized devices. Processing stays on this computer. Stop disconnects clients and removes the token.

Use another computer

Offload to Wi-Fi

Send prompts and audio to your chosen Wi-Fi computer. Its models do the work.

Computer API addressAccess token

Available in the local API server.

Tutorial ↗ · API guide ↗

Local storage

Approximate storage used by this site: Checking…

Models and creations stay in this browser’s site storage. The browser controls where files live on disk; this page cannot choose a folder.

Downloaded models

Remove cached model files. You can download them again later.

Creation history

Remove all saved voice recordings and captions from this browser.

Clearing model files does not remove your creations.