1. Paste the script

Choose a language and paste your text. Check the line breaks before generating: each line can be edited, moved, or given its own speaker.

2. Listen to the lines

The voice model downloads on first use, then makes speech on your device. Fix a misread line and generate that line again.

3. Download the MP3

Export one joined MP3 for sharing. If you need separate clips for a video editor, choose the ZIP export instead.

Why can we offer this for free?

The work happens on your device. After the voice model downloads, your browser turns each line into speech instead of sending the script to a paid speech server. That avoids a per-request speech bill.

The website and model downloads still need hosting and maintenance. Free means you can use the local editor and export without an account or a speech credit meter. It does not mean running the service has no cost.

How does the browser make the audio?

On first use, the browser downloads model files from Hugging Face. Most supported languages use Supertonic 3. Hebrew uses BlueTTS 2.5 and ReNikud to help turn unpointed text into pronunciation data.

A browser worker runs the ONNX models with ONNX Runtime Web and WebAssembly. It takes the text, voice, and speed for one line and returns audio samples. The editor keeps those samples in the browser, so you can regenerate a corrected line without making the others again.

For an MP3 download, lamejs encodes the samples in the browser. The editor can also join lines with your chosen pauses, write a WAV file, package separate MP3 clips with JSZip, or make SRT cues from the generated line lengths.

See the model sources and licenses for details on the tools used here and the terms that apply to them.

Try a small script

For a three-sentence intro, listen to the name in sentence two before exporting. If it sounds wrong, change only that line.

When this fits

Short scripts, video narration, and a quick audio draft you can revise sentence by sentence.

Limit: The first model download is large, and generation speed depends on your device and available memory. The model may be cached for later visits, but clearing browser data can require another download. Check pronunciation and the final file before publishing.