BRev Text to speech — 32 languages, free & private
Text to speech in 32 languages — device voices speak instantly; a neural AI voice downloads once, runs on your device and saves a WAV.
This text to speech tool has two engines. Your device’s voices use the browser's built-in speech synthesiser: nothing to download, a Speed slider from 0.5× to 2× and Pitch from 0.5 to 1.8, and a list of whatever voices your operating system has installed. They play through your speakers only, because browsers do not hand that audio back to the page, so there is nothing to save. The downloadable AI voices are 53 Piper neural voices covering 32 languages, from Arabic and Chinese to Welsh and Swahili. The first time you use one, its model file, between 21 MB and 76 MB and shown in the picker beforehand, is fetched from Hugging Face and stored in this browser; later uses load it from storage. Speech is then generated on your device with WebAssembly, in chunks of up to about 380 characters split at sentence ends, and joined into a mono 16-bit WAV you can play or download. The text box takes up to 20,000 characters for a neural voice.
What you can do
- Two engines side by side: Your device’s voices (instant, 0 MB) and Downloadable AI voices (32 languages, saves a file)
- 53 Piper neural voices in 32 languages, including Arabic, Farsi, Chinese, Ukrainian and Swahili; English has four, two British and two American
- The download size of every AI voice shown in the picker before you commit: 21 or 28 MB for Compact voices, 63 to 76 MB for most Standard ones
- Voices are kept in the browser's private file storage after the first use, listed under Downloaded voices with Remove and Remove all
- Download .wav: mono, 16-bit, at the voice's own sample rate, named from the first words of your text, with 120 ms of silence between chunks
- Speed (0.5× to 2×), Pitch (0.5 to 1.8), Pause, Resume and Stop for device voices
- Live character, word and speaking-time counts, and a text box that lays out Arabic and Farsi right to left as you type
How to use text to speech
- Type or paste into “Your text”The counter shows characters out of 20,000, the word count and an estimate of speaking time at 180 words a minute. Press Sample to load a test sentence.
- For instant playback, choose “Your device’s voices” and press “Speak”Pick a Voice from your operating system's list and set Speed and Pitch. Pause and Stop appear while it talks; this audio plays only and cannot be saved.
- For a file, choose “Downloadable AI voices”Pick a Language and a Voice. The Download box shows that voice's size, or Ready if it is already cached on this device.
- Press “Generate speech”On first use the voice downloads with a progress bar, then the text is spoken chunk by chunk. When it finishes, the audio plays in the page.
- Press “Download .wav”The file is named from the first words of your text. If you need an MP3, the Video & audio converter's Extract audio option converts it in the browser.
FAQ
Can I download the speech as an MP3 or WAV file?
The downloadable AI voices save a WAV file: mono, 16-bit, at the voice's own sample rate. There is no MP3 button here, but the Video & audio converter on this site turns the WAV into a 192 kbps MP3 with its Extract audio option, also in your browser. Device voices cannot be saved at all, because the browser plays them without giving the page the audio.
Which languages does it support?
The neural voices cover 32 languages: Arabic, Catalan, Chinese, Czech, Danish, Dutch, English, Farsi, Finnish, French, Georgian, German, Greek, Hungarian, Icelandic, Italian, Kazakh, Luxembourgish, Norwegian, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swahili, Swedish, Turkish, Ukrainian, Vietnamese and Welsh. Japanese, Korean and Hindi are not in that list. Device voices cover whichever languages your operating system has installed, grouped by language in the Voice menu.
How big are the AI voice downloads, and do I download them every time?
Each voice is one model file. The picker shows its size: 21 or 28 MB for Compact voices, 63 or 64 MB for most others and 76 MB for the two Icelandic voices. It is fetched from Hugging Face the first time and kept in the browser's origin private file system, so later uses load it from storage; the first use also loads the speech runtime from this site, whose phonemiser data alone is 18 MB. Remove or Remove all under Downloaded voices frees the space.
Is my text sent anywhere?
Not with a downloadable AI voice. Its only network requests are for the voice model, which is addressed by voice name, and the speech runtime, and your text is turned into audio on your device. Device voices come from your browser and operating system, and browsers mark some voices as not local, meaning an online service may produce the audio; for text that must stay on the device, use an AI voice.
How long a text can I turn into speech?
Up to 20,000 characters; Generate speech is disabled above that. The text is split at sentence ends, including the Arabic question mark ؟ and the Chinese full stop 。, into chunks of up to about 380 characters, and the progress bar moves one chunk at a time. Cancel stops before the next chunk starts.
Does it read Arabic?
Yes. The neural list has one Arabic voice, Kareem, recorded for Jordan, which is a 63 MB download. The text box sets its direction automatically, so Arabic lays out right to left as you type. Whether a device voice can read Arabic depends on the voices your operating system has installed.
Why do the device voices differ from one computer to another?
They are whatever your operating system and browser provide, so the list changes from one device to the next. The tool re-reads the list when the browser fires its voiceschanged event, because many browsers report no voices on the first call. If a device has none, the page says so, and the downloadable AI voices still work.