Text to speech you can download as MP3
This text to speech MP3 maker turns a script, an article or a few lines into a natural-sounding voice-over and lets you download it as an MP3 (or WAV), with matching SRT subtitles. Choose from six English voices — American and British, female and male — or four Hindi voices, set the speed and the pauses, and press Make audio.
The speech is generated on your own device by Kokoro-82M, an open neural text-to-speech model, so your text never leaves your computer or phone and there is no sign-up, word limit per day or watermark. The first time, the model (88 MB) is downloaded once and kept by your browser. It is a free voice over generator for YouTube videos, reels, presentations, e-learning and audiobooks drafts.
How to use the Text to Speech MP3
- Type or paste your text. Up to 20,000 characters can be made in one go; long texts are read sentence by sentence.
- Pick a voice: English (US), English (UK) or Hindi. Use the Hindi example button to try a Hindi voice-over.
- Adjust the speed and the pause between sentences. Blank lines between paragraphs add a longer pause.
- Press Make audio. The first run downloads the voice model; then each part is spoken on your device, with progress shown.
- Listen, then press Download MP3, Download WAV or Subtitles (SRT).
Text to Speech MP3 features
Real downloadable audio
MP3 at 64–192 kbps or 16-bit WAV — unlike the read-aloud voices built into browsers, which cannot be saved.
English and Hindi voices
Ten Kokoro voices: American and British English, and Hindi female and male voices for Hindi voice-overs.
Reads numbers properly
Dates, years, money ($, £, €, ₹), percentages and times are spoken in words — in Hindi for Hindi voices.
Pronunciation fixes
Tell the voice how to say a name or brand (“Nguyen = win”), or give exact phonemes.
Subtitles included
An SRT file with one caption per sentence, timed to the audio, for video editors and YouTube.
Long texts
Sentence-by-sentence generation with progress, Stop, and re-use of finished parts when you edit and remake.
How it works
Your text is first split into sentences, and numbers, money and abbreviations are written out as words. English words are looked up in the misaki pronunciation dictionary (the lexicon Kokoro was trained with, about 90,000 words, Apache-2.0); plurals, past tenses and -ing forms are built from it, acronyms are spelled, and the rare word that is missing is read by letter-to-sound rules and listed under Pronunciation fixes. Hindi in Devanagari is read with the rules of the script, including the silent final “a”.
Each sentence (split at commas if it is very long) goes to the Kokoro-82M model, which runs in a background worker with WebAssembly through transformers.js. Silence at the ends of each part is trimmed, your chosen pauses are inserted, and the whole voice-over is levelled to a −1 dB peak before it is encoded as MP3 with the LAME encoder.
Download size and speed
The voice runs on a single processor core in WebAssembly, so making the audio typically takes two to four times as long as the finished speech on a laptop, and longer on phones. A 1,000-word script (about 7 minutes of audio) can take 15–30 minutes: keep the tab open, and use Stop if you need to — finished parts are kept and reused.
| Item | Size | When |
|---|---|---|
| Kokoro-82M voice model (8-bit) | 88 MB | First Make audio, then cached |
| Voice style file | about 0.5 MB per voice | First use of each voice |
| English dictionary (US or UK) | under 1 MB | First English voice-over |
| MP3 encoder | about 0.3 MB | First MP3 download |
Text to speech MP3 or read aloud?
To simply listen to text, the Text to Speech reader uses your device’s built-in voices and highlights each word, but browsers do not let a page record those voices. This page brings its own voice model so the result can be saved. For a spoken name over a ringtone tune, see the Name Ringtone Maker; to join voice-over parts or add music, use the Audio Joiner.
- The voices are synthetic. Say so where it matters, and don’t use them to imitate a real person.
- Hindi voices handle everyday Hindi well; very formal Sanskrit-heavy words and some names may need a fix.
- Very long or unusual sentences can sound flat — shorter sentences with commas give a livelier read.
Text to Speech MP3 FAQ
How do I download text to speech as an MP3?
Paste your text, choose a voice, press Make audio and then Download MP3. The file is made in your browser; nothing is sent to a server.
Is this text to speech MP3 maker free?
Yes. There is no sign-up, no daily limit and no watermark. The only cost is the one-time 88 MB model download.
Can I make a Hindi voice over?
Yes. Choose one of the four Hindi voices and write in Devanagari (or Hindi in English letters). Numbers are read in Hindi.
Can I use the MP3 in my YouTube videos?
The voice model is released under the Apache-2.0 licence, which allows commercial use, and the audio is generated by you. Check the rules of the platform you publish on, and label synthetic voices where required.
Why does a word sound wrong?
Names, brands and new words may not be in the dictionary. Open Pronunciation fixes and write how the word sounds, for example “Nguyen = win”, then make the audio again.
Does it work offline?
After the first run, the model and dictionary are cached by your browser, so later voice-overs can be made without downloading them again.
Why not use the voices built into my browser?
Browsers play those voices straight to the speakers and do not give web pages the audio, so they cannot be saved as MP3.