Text to Speech

Tell it how to sound

The feature to actually use, and the one most free text-to-speech tools do not have.

There are three separate controls for the delivery, on top of the voice, and they stack. Most free text-to-speech tools give you none of them.

Vibe / Emotion is a row of ready-made presets — Neutral, Cheerful, Calm, Serious, Excited, Storyteller, Newscaster and Whisper. This is the quickest win on the page and the one to try first. Storyteller and Newscaster in particular produce noticeably different readings of the same sentence: one is expressive with dramatic pauses, the other clear and formal.

Tone is a five-step slider from Very soft through Balanced to Very energetic. It adjusts how much emphasis and intonation the delivery carries, and it works alongside the vibe rather than replacing it — Storyteller at Very soft is a bedtime reading, Storyteller at Very energetic is an adventure.

Custom instructions is the real power control: a free-text field where you describe the delivery in your own words. For example:

"Speak with a slight French accent, pause before key points." · "Warm and unhurried, like a documentary narrator." · "Bright and quick, as if reading good news." · "Calm and clear, for a safety announcement."

The same voice reading the same sentence under two different instructions produces noticeably different audio — different pacing, different emphasis, a different emotional register. That is far more control than choosing between sixty voice names, because the voice sets the timbre while the instruction sets the performance.

What works in an instruction: a pace ("unhurried", "brisk"), a mood ("warm", "serious", "playful"), a context ("like a news reader", "like an audiobook narrator", "as if explaining to a child"), and how to treat emphasis. Keep it short — a phrase or two is more effective than a paragraph.

A speed control also appears where the underlying model supports one. The interface shows what is actually available rather than offering a control that does nothing, which is the right way round.

A short voice list, and why that is enough

Being straight about a comparison that is easy to make.

Several free text-to-speech sites advertise fifty or sixty voices. This one offers under a dozen, each labelled with what it actually sounds like — bright, deep and informative, firm and confident, breezy, youthful — rather than just a name. In practice the gap matters much less than it sounds, for two reasons.

A long voice list is mostly the same few voices in different accents and languages. What you are choosing between for any given piece of work is usually three or four plausible options, and a short list of distinct ones covers that comfortably.

And the style controls do more to the output than the voice choice does. A warm instruction on a neutral voice beats a "warm voice" read flatly. So the controls that matter here are the ones this tool has, and the voice list is the part where it concedes.

The practical advice: try three voices on one sentence before committing to a long piece of text. Voices differ more in how they handle specific sounds and pacing than a name or a description can convey, and thirty seconds of testing saves redoing a five-minute recording.

Preparing the text is most of the work

The difference between speech that sounds natural and speech that sounds like a machine is usually in the text, not the settings.

Punctuate for the ear. Commas and full stops become pauses, so punctuation is your pacing control. A long sentence without commas comes out breathless. Adding a comma where you would naturally draw breath fixes it immediately.

Spell out anything ambiguous. "Dr." may be read as "doctor" or "drive". "1996" may be "nineteen ninety-six" or "one thousand nine hundred and ninety-six". "St" could be street or saint. If a word must be read a particular way, write it that way — this is the single most common cause of a re-record.

Expand abbreviations and units unless you want them spelled out. "approx." and "3kg" and "e.g." all read better as words.

Break up walls of text. Paragraph breaks and shorter sentences give the delivery somewhere to breathe, and they help the model place emphasis sensibly.

Read your text aloud yourself first. Anything you stumble over, the synthesiser will stumble over too — and it is much faster to fix the sentence than to fight the settings.

What it is good for, and what it is not

Good uses: a voiceover for a video, a presentation or a tutorial · listening to your own writing back, which is the best proofreading technique there is and catches awkward sentences nothing else does · an audio version of an article, a set of notes or study material · accessibility, where someone would rather listen than read · practising pronunciation in another language · a quick announcement, prompt or phone message.

Where it still falls short:

Long-form narration. Modern synthesis is very good for a minute and gradually less convincing over twenty, because it lacks the long-range variation a human narrator brings. For an audiobook, expect to break it into sections and accept it will not match a professional reading.

Emotional range. The instructions help considerably, but genuine emotional performance — grief, comic timing, building tension — is still beyond it, and forcing it usually sounds worse than a neutral read.

Names and specialist terms. Personal names, place names, medical and technical vocabulary are frequently mispronounced. Listen before you publish, and respell anything wrong phonetically.

Anything where a listener would object to a synthetic voice — and be aware that using a synthetic voice in a commercial or public context increasingly invites disclosure, and that impersonating a real person's voice is a serious matter both ethically and, in a growing number of places, legally.

Practical notes

The audio plays in the page so you can check it before downloading, and the download format depends on the model in use — commonly an MP3, which plays and imports anywhere.

You can paste up to 12,000 characters — roughly 2,000 words, or fifteen minutes of speech — which is far more generous than most free tools, several of which stop at a few hundred. A live character and word count sits under the box.

Even so, work in short passages. Generate a paragraph, listen, adjust, then move on. It is faster than producing a long piece and discovering a mispronunciation two minutes in, and it lets you use a different vibe for different sections.

There is a limit of 15 generations an hour. Worth planning around if you are producing something substantial, and worth knowing that some competitors advertise unlimited use — though generally with fewer controls, or a watermark, or an account.

Unlike the file tools on this site, your text is sent to a server to be spoken, so do not paste anything confidential. There is no account needed.

Related tools

Text Summarizer shortens a long piece before you narrate it, which usually improves the result. Word Counter gives you a reading-time estimate, useful for matching a script to a video length. Text Case Converter tidies text that arrived in inconsistent capitalisation, which affects how it is read.