AI Voices

How SpeakScribe voices work

Type or paste a script, pick one of 361 voices in 75 languages, generate, and download the audio. This page covers what is the same for every voice — the two voice types, file formats, limits, subtitles, long scripts and how long your audio will be. Each voice page keeps what is specific to that voice: its measured pitch and pace and the closest alternatives.

From text to audio

  1. 1

    Pick a voice

    Browse voices by language. Every voice page plays a ready-made sample — it costs nothing and needs no account.

  2. 2

    Paste your script

    Type on the voice page or open Voice Studio for longer texts. Right-to-left scripts such as Arabic, Persian and Urdu are handled as written.

  3. 3

    Generate

    A short text is usually ready in 5–20 seconds. Long scripts are rendered in parts, with progress shown, and joined into one file.

  4. 4

    Download

    Save the MP3 or WAV. On paid plans, SRT and VTT subtitles timed to the speech come with it.

Two kinds of voices: Studio and Global

Every voice page shows a Studio or Global label. Both are free to use; they differ in where they run and what you can adjust.

StudioGlobal
Voices 41 voices in 6 languages 320 voices in 75 languages
Runs on SpeakScribe’s own servers, using the open Kokoro neural model Microsoft neural text-to-speech, online
Speed Adjustable from 0.5× to 2× in Voice Studio Natural pace; slow down, speed up or change pitch for parts of a script with SSML
Download WAV; MP3 for long scripts MP3
Best when You want to slow a lesson down or speed a short clip up You need a regional accent or a language Studio does not cover

Limits by plan

The same limits apply to every voice. A render is one generated audio file; playing a voice sample is never counted.

PlanRenders Chars / renderSubtitles Files kept
No account 3 5,000 ——
Free 10 / mo 5,000 — 7d
Starter 100 / mo 30,000 SRT + VTT 30d
Pro 500 / mo 70,000 SRT + VTT 90d
Studio 1,500 / mo 100,000 SRT + VTT 180d
Agency Unlimited 150,000 SRT + VTT 365d

Guest renders count toward the free allowance: create a free account to use the rest. Prices are on the pricing page.

How long will my audio be?

It depends mostly on the language and its alphabet. Below is the median pace of all voices in each language, measured on 30 October 2026 while every voice read the same ~15-second passage (characters counted with spaces, as in the plan limits). Each voice page shows that voice’s own pace.

Afrikaans 1,000 chars ≈ 1 min 3 s
Albanian 1,000 chars ≈ 1 min 1 s
Amharic 1,000 chars ≈ 1 min 31 s
Arabic 1,000 chars ≈ 1 min 45 s
Azerbaijani 1,000 chars ≈ 1 min 7 s
Bengali 1,000 chars ≈ 1 min 12 s
Bosnian 1,000 chars ≈ 1 min 8 s
Bulgarian 1,000 chars ≈ 1 min 9 s
Burmese 1,000 chars ≈ 1 min 3 s
Catalan 1,000 chars ≈ 1 min 12 s
Chinese 1,000 chars ≈ 3 min 30 s
Croatian 1,000 chars ≈ 1 min 10 s
Czech 1,000 chars ≈ 1 min 12 s
Danish 1,000 chars ≈ 1 min
Dutch 1,000 chars ≈ 1 min 7 s
English 1,000 chars ≈ 1 min 3 s
Estonian 1,000 chars ≈ 1 min 6 s
Filipino 1,000 chars ≈ 1 min 2 s
Finnish 1,000 chars ≈ 1 min 12 s
French 1,000 chars ≈ 56 s
Galician 1,000 chars ≈ 1 min 2 s
Georgian 1,000 chars ≈ 1 min 6 s
German 1,000 chars ≈ 1 min 4 s
Greek 1,000 chars ≈ 56 s
Gujarati 1,000 chars ≈ 1 min 30 s
Hebrew 1,000 chars ≈ 1 min 29 s
Hindi 1,000 chars ≈ 1 min 18 s
Hungarian 1,000 chars ≈ 1 min 14 s
Icelandic 1,000 chars ≈ 1 min 1 s
Indonesian 1,000 chars ≈ 1 min 12 s
Inuktitut 1,000 chars ≈ 2 min 7 s
Irish 1,000 chars ≈ 1 min 5 s
Italian 1,000 chars ≈ 1 min 2 s
Japanese 1,000 chars ≈ 2 min 43 s
Javanese 1,000 chars ≈ 1 min 12 s
Kannada 1,000 chars ≈ 1 min 12 s
Kazakh 1,000 chars ≈ 1 min 7 s
Khmer 1,000 chars ≈ 1 min 14 s
Korean 1,000 chars ≈ 2 min 29 s
Lao 1,000 chars ≈ 1 min 22 s
Latvian 1,000 chars ≈ 1 min 8 s
Lithuanian 1,000 chars ≈ 1 min 10 s
Macedonian 1,000 chars ≈ 1 min 3 s
Malay 1,000 chars ≈ 1 min 11 s
Malayalam 1,000 chars ≈ 59 s
Maltese 1,000 chars ≈ 1 min
Marathi 1,000 chars ≈ 1 min 26 s
Mongolian 1,000 chars ≈ 58 s
Nepali 1,000 chars ≈ 1 min 16 s
Norwegian 1,000 chars ≈ 1 min 3 s
Pashto 1,000 chars ≈ 1 min 6 s
Persian 1,000 chars ≈ 1 min 16 s
Polish 1,000 chars ≈ 1 min 3 s
Portuguese 1,000 chars ≈ 1 min 1 s
Romanian 1,000 chars ≈ 1 min 5 s
Russian 1,000 chars ≈ 1 min 10 s
Serbian 1,000 chars ≈ 1 min 10 s
Sinhala 1,000 chars ≈ 1 min 8 s
Slovak 1,000 chars ≈ 1 min 12 s
Slovenian 1,000 chars ≈ 1 min 12 s
Somali 1,000 chars ≈ 1 min 4 s
Spanish 1,000 chars ≈ 1 min 4 s
Sundanese 1,000 chars ≈ 1 min 14 s
Swahili 1,000 chars ≈ 1 min 13 s
Swedish 1,000 chars ≈ 1 min 12 s
Tamil 1,000 chars ≈ 1 min 16 s
Telugu 1,000 chars ≈ 1 min 20 s
Thai 1,000 chars ≈ 58 s
Turkish 1,000 chars ≈ 1 min 10 s
Ukrainian 1,000 chars ≈ 1 min 13 s
Urdu 1,000 chars ≈ 1 min 17 s
Uzbek 1,000 chars ≈ 1 min 5 s
Vietnamese 1,000 chars ≈ 1 min 1 s
Welsh 1,000 chars ≈ 1 min 7 s
Zulu 1,000 chars ≈ 1 min 13 s

Scripts with many numbers, names or long pauses run longer. Chinese, Japanese and Korean pack more speech into each character, so they need fewer characters per minute. How we measure

Long scripts

Scripts over about 2,000 characters are split into parts, rendered one after another in the same voice and joined into a single file. Chinese, Japanese and Korean characters count three times toward that threshold. One long script per account runs at a time; the page shows its progress.

Subtitles

On paid plans each render also produces SRT and VTT subtitles timed to the voice, ready for YouTube or a video editor. How accurate the timing is

Your files

Audio you generate is private to your account and listed in your history until it expires — 7 days on the free plan, longer on paid plans (see the table above). Download anything you want to keep.

Making YouTube videos with SpeakScribe voices

The workflow most creators follow, with the limits that apply at each step.

  1. Write the script for listening. Short sentences, numbers written the way you say them, a blank line where the scene changes. English voices read about 16 characters a second (median), so a 10-minute video needs roughly 9,600 characters.
  2. Choose a voice by its numbers. Each voice page lists measured pitch, intonation range and pace. A wider range sounds livelier, a narrower one steadier.
  3. Preview a few lines to catch names, numbers or words read the wrong way, and fix them in the text.
  4. Render. A free account renders up to 5,000 characters at a time; paid plans render longer scripts in one go (see limits by plan) and join them into a single file.
  5. Add captions. Paid plans export SRT and VTT subtitles timed from the voice (average offset 26–97 ms in our test); upload the SRT to YouTube Studio with the video.
  6. Level the audio. AI voices come out quieter than most YouTube videos; normalize the voice before upload (how, with measurements).

Where you can use the audio

Studio voices: any purpose, including paid advertising, broadcast and work you deliver to clients. Global voices: your own content, including monetized videos on your channel; for paid ads, broadcast or client work use a Studio voice. No attribution needed (Terms, section 6). Whether a channel earns money is decided by YouTube under its own content policies, which favour original videos.

Two or more voices in one video

For interviews, debates or a two-host format, Multi-voice gives each line its own speaker and returns one MP3 with every voice levelled to the same loudness. Free accounts can try 2 speakers (3 dialogues a month, up to 1,500 characters); paid plans allow 4 to 10.

Find a voice

Frequently asked questions

Yes. The Free plan has no end date and needs no card: 10 voiceovers every month, each up to 5,000 characters, with every voice included. You can also make 3 without an account (up to 1,500 characters each). There is no trial period that runs out — paid plans only add more and longer voiceovers, plus subtitles.
No. Free voiceovers are clean audio of your script — no watermark, no spoken intro, no logo sound — the same files paid plans get.
No. Playing a voice sample never needs an account, and you can make 3 voiceovers as a guest. Creating a free account unlocks the rest of the 10 monthly voiceovers and keeps your files in your history.
Yes, in the browser on Android or iPhone — type or paste your text, generate, and save the MP3 to your phone. You can add SpeakScribe to your home screen so it opens like an app.
Studio voices have a speed setting from 0.5× to 2× in Voice Studio — around 0.8× suits lessons, language practice and readers who prefer a slower pace. Global voices read at their natural pace, but you can wrap parts of a script in SSML — <prosody rate="slow"> or a <break> — to change speed, pitch and pauses.
Start a new line or paragraph for a longer pause; commas and full stops give shorter ones. For an exact pause, add an SSML break such as <break time="2s"/> where you want the silence.
Write it the way it should be said. Dates: "12 May 2026" or "the twelfth of May" instead of 12/05/2026, which can be read in different orders. Abbreviations: "U.S.A." to spell it out, "NASA" to say it as a word. Names: spell them by sound, and for Hindi, Urdu or Arabic names type them in their own script. Remove emojis and symbols you do not want read aloud.
Global voices give you an MP3. Studio voices give you a WAV for normal scripts and an MP3 for long scripts, which are joined into one file. A ten-minute MP3 voiceover is a few megabytes — small enough to send or upload anywhere.
Download the MP3 or WAV and import it like any audio file — add it from your device's files in CapCut, upload it to your media in Canva, or drag it onto an audio track in Premiere Pro. On paid plans, import the SRT file too and the captions line up with the voice.
Yes, on paid plans every render also produces SRT and VTT subtitle files timed to the speech. Free and guest renders give audio only.
It turns any text you paste into an audio file you can play anywhere, at a slower pace with Studio voices. It is not a screen reader — it reads the text you give it, not the screen.
Multilingual voices can: they detect the language of the text they are given (very short lines are sometimes detected wrongly). Other voices are made for one language, so text in a different language or alphabet usually comes out with the wrong pronunciation. Pick a voice of the language you are writing in.
Studio voices run on SpeakScribe's own servers with the open Kokoro model and have a speed slider; Global voices are Microsoft neural voices — the same family Edge's Read Aloud uses — and cover far more languages and regional accents. Here you can download either as a file.
Not in one render — each voiceover uses one voice. Make each speaker's lines separately and place them one after another in your video or audio editor.
No. There is no emotion selector and no sound effects. Choose a voice by its measured character — calmer, livelier, deeper — and use punctuation or SSML emphasis to shape the delivery.
No. Voice cloning is not offered, there is no public API, and each voiceover is made one at a time in the browser.
The character counter shows when you are over. Split the script into several renders, or use a plan with a higher per-render limit. Scripts over about 2,000 characters are processed in parts and come back as one file.

Try a voice on your own script

Open Voice Studio, paste your text and pick any of the 361 voices.

Open Voice Studio