Text to Speech

It depends which voice you pick, and the tool tells you which is active. A device voice runs entirely in your browser or operating system, so nothing leaves your computer. A neural voice sends your text securely to the speech provider to generate the audio — the page states this before you generate, and no "no upload" claim is shown while a cloud voice is selected.

English, Bengali (বাংলা) and Hindi (हिन्दी) are the three the tool is tested against most carefully, alongside Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Urdu, Arabic, Spanish, French, German, Portuguese, Russian, Japanese, Korean and Chinese among others. Device voices depend on what your operating system has installed, so the voice list shows exactly what is available to you rather than a fixed promise.

With a neural voice, yes — you get a standard WAV file with no watermark and no spoken attribution. Device voices cannot be downloaded, because browsers do not expose system speech synthesis for recording. Rather than show a download button that fails, the tool hides export entirely for those voices and says why.

Yes, and the voices are tested with Bengali and Devanagari script, including the danda punctuation that Indic scripts use. Two things worth knowing: pick a voice that matches your text, because an English voice reading Bengali produces nonsense; and if your text mixes scripts — an English term inside a Bengali sentence, for example — the tool warns you and suggests splitting it into per-language segments.

Device voices support both, from 0.5× to 2× speed and a pitch range either side of neutral. Neural voices set their own delivery and expose neither, so those controls are hidden rather than shown doing nothing. Presets like Narration, Podcast and Proofread adjust whatever the selected voice actually supports.

Yes. Text is split at paragraph and sentence boundaries — never mid-word and never inside a Bengali cluster or an emoji — synthesised in pieces, and joined back into one file with a short fade at each seam so there are no clicks. You can also see the segment progress and cancel partway without losing your text.

Yes, with segments. Split your script into parts, give each one its own language and voice, and generate them together. That is the practical way to handle bilingual dialogue, or a narrator and a character in different languages.

Rights depend on the voice you selected and its provider's terms, so the tool does not promise blanket commercial use. Check the terms for the specific voice before using it in a commercial project.

Device voices often do, since they use voices installed on your machine — though some browsers still need a connection for certain engines. Neural voices require an internet connection by definition. The tool does not claim offline support it cannot verify.