Recentech StudioRecentech Studio

Free Text to Speech. Convert text to natural speech online for free. Multiple AI voices and languages, instant audio playback — great for learning and accessibility.

Text-to-Speech

Convert text to natural speech with the Kokoro-82M neural model running 100% in your browser — 53 real voices across 9 languages, with a one-time ~86 MB download only when you ask for it.

Text to Convert
Enter the text you want to convert to speech (max 5000 characters)
~0 sec0 words
0 / 5000 characters
Try an example:
Voice Engineruns on your device
HD neural voices
86 MB once

Kokoro-82M neural model — most natural quality and real WAV export. 53 voices across 9 languages.

Checking local model cache…

Speed

⚠️ AI may produce inaccurate information. Please verify important details.

Key Features & Benefits

HD Neural Voices — Kokoro-82M

53 real neural voices across 9 languages: English (US/UK), Spanish, French, Italian, Portuguese (Brazil), Hindi, Japanese, and Chinese. The model runs entirely in your browser — no server, no API.

One-Time Download, Only When You Ask

The voice model (~86 MB) is never downloaded automatically. You press the download button, it saves into your browser cache once, and from then on every generation happens on your device.

100% Private — On-Device AI

Everything runs on your device. Your text never touches a server — no API calls, no tracking, no data collection. What you type stays with you.

Download Real Audio (WAV)

Export the actual generated sound as a WAV file — ready for video voiceovers, podcasts, or study material. The audio is truly generated on your hardware.

Real Speech Markup

Only markup that genuinely works: [pause:1s] inserts a real second of silence, [slow] and [fast] really change the pace of that section. Long texts are split into sentences and merged into one smooth file.

Free — No Credits, No API Keys

Like the other on-device labs, this tool costs 0 credits. The neural model is Apache-2.0 licensed and runs on your hardware — no sign-ups, no per-character billing, no server costs.

Full Playback Controls

Play, pause, and stop with a progress bar that tracks the real audio position of the generated file.

Works Offline After Setup

Once the model is cached in your browser, the tool works offline. You can remove the cached model anytime with one click from the Voice Engine card.

Frequently Asked Questions

How does Text-to-Speech work?
It runs the Kokoro-82M neural model inside your browser via WebAssembly. Press 'Download voice model' once (~86 MB, stored in your browser cache) — from then on every generation happens on your device, with real WAV export.
Is my text data stored or sent anywhere?
No. Everything processes locally in your browser. Your text is never sent to any server, never stored, and disappears when you close the page.
Why does the tool need a one-time download?
A neural voice model is a large file — about 86 MB. It is only downloaded when you press the download button; nothing downloads automatically or in the background. It is saved in your browser cache (nothing is installed on your computer), works offline afterwards, and you can remove it anytime from the Voice Engine card.
How do I remove the model from my device?
Open the Voice Engine card and click 'Remove cached model'. The model files are deleted from your browser cache immediately — nothing remains on your computer.
How fast is generation?
Faster than real time on computers with WebGPU; on CPU-only devices a 10-second clip typically takes 10–60 seconds, with a live progress indicator throughout.
Can I download the audio?
Yes — every generation can be saved as a real 24 kHz WAV file, ready to use anywhere.
Which languages are supported?
Kokoro-82M supports 9 language groups: English (US/UK) at the highest quality, plus Spanish, French, Italian, Portuguese (Brazil), Hindi, Japanese, and Chinese — 53 real voices grouped by language. This tool does not read or use the voices installed on your device.
Does this cost credits?
No. Like the other on-device labs, this tool is completely free: it runs on your hardware, so there are no server costs to cover — no credits, no API keys, no usage limits.