Ready-made, invented or cloned
6 ready-made voices
From a warm audiobook narrator to a serious newsreader to a spooky horror voice. Just click, listen to a sample and get started.
Describe a voice
“Deep, gravelly pirate voice, loud and laughing”: describe age, tone, mood and pace, and the AI invents a voice that has never existed before.
Clone your own voice
Read a sentence into your microphone or upload a recording. Just 5 to 10 seconds is enough for the AI to speak any text in your voice.
Here's how the ready-made voices sound
All samples are unedited, exactly as they come out of the model.
Three steps to a finished MP3
- Paste your textAn article, email, study notes or a story, up to 3000 words.
- Choose a voiceA ready-made voice, your own description or your cloned voice, plus the language of the text.
- Listen & downloadShort texts are done in seconds, 300 words in about a minute. Then save it as an MP3.
What people use it for
Listen to long texts on the go
Listen to articles, reports or emails while cooking, commuting or working out instead of reading them on screen.
Learning & languages
Review study material as audio, or have texts read aloud in English, French, Spanish and seven other languages.
Easier reading
Help with dyslexia, tired eyes or visual impairment: listen instead of deciphering.
Videos & presentations
Voiceovers for YouTube, Reels, explainer videos or slides, without ever stepping up to the mic yourself.
Stories & audio dramas
Bedtime stories in a grandpa voice, horror in a whisper, or your own characters created from a description.
Greetings in your voice
A birthday greeting or voice message in your cloned voice, even in a language you don't speak.
Our daily podcast is voiced with the same technology
Every morning, “Tokenstrom – AI News of the Day” sums up the most important AI news, spoken by this very model on our own GPU. Listen in and hear how natural ten minutes straight can sound.
Anthropic's Record IPO, an OpenAI Agent Goes Rogue in Training, and Tencent's Hy4 | November 20, 2026
Subscribe on Spotify, Amazon Music, Deezer and everywhere else with this feed: https://tokenstrom.com/podcast/feed.xml
How it differs from the big paid services
| Tokenstrom | typical paid services | |
|---|---|---|
| Price | free | free tier, then subscription |
| Account required | no | yes |
| Voice cloning | yes, free | usually subscription only |
| Your data | voice sample deleted immediately, audio after 1 hour | often stored, sometimes used for training |
| Model | open source (Apache 2.0), our own GPU in Germany | closed, cloud |
To be honest: the top services sometimes still sound a touch better on very long texts and emotional delivery. For everyday use, Qwen3-TTS is more than enough.
More on AI voices and speech models
FAQ
Is text-to-speech really free?
Yes. No subscription, no credits and no sign-up. We cover the cost of electricity and the GPU so you can try AI for yourself.
How long can the text be?
Up to 3000 words per run. For longer texts, just split them into several parts.
Which languages are supported?
German, English, French, Spanish, Italian, Portuguese, Russian, Chinese, Japanese and Korean. Your cloned voice can even speak languages you don't speak yourself.
How do I clone my voice?
In the “Clone your own voice” tab, you can record 5 to 20 seconds right in your browser or upload an audio file. You'll get the best results in a quiet room without echo, and if you enter the exact words you said in the recording.
Is voice cloning allowed?
Yes, with your own voice or with the person's explicit permission. Cloning someone else's voice without consent, or using it to deceive people, isn't allowed and is often a crime.
What happens to my voice sample and my text?
Your voice sample is deleted right after the speech is generated. The finished audio is deleted after one hour, or right away if you click the button. Nothing is published or used for training. Everything runs on our own GPU in Germany.
What AI powers this?
Qwen3-TTS with 1.7 billion parameters from Alibaba's Qwen team, an open model under the Apache 2.0 license. We run it on an NVIDIA Tesla P40.
Why do I sometimes have to wait?
There's exactly one GPU, and Listen, Wallpaper and Podcast all share it. If it's busy, you'll join a short queue.
Just give it a try.
No account, no credit card, no watermark subscription. Paste text, pick a voice, listen.
Go to the Listen tool