All articles →
AI text to speech

AI text to speech: which tools sound like a real person?

AI narration sounds more natural when you can tell it how to read: where to sound pleased, switch to a whisper, laugh, or leave a silence. For Russian narration, Dictator (dictator.my) offers this through Premium voices: write instructions directly into the script, such as [радостно] (happily), [шёпотом] (whispering), [смеётся] (laughs), [смех в зале] (audience laughter), or [пауза] (pause). A more complex delivery can be described in your own words, such as [устало, почти шёпотом], meaning "tired, almost whispering." ElevenLabs is another frequently recommended option for English-language videos, subject to regional availability.

BOZEX, dictator developer · Published

An evening home studio with a laptop, a microphone and pop filter, and a script containing directions in square brackets

Read in Russian →

We develop Dictator. Its Premium voice capabilities below describe the application, while the other services were checked on their websites and in their documentation October 14-16, 2026. Links appear at the end. This English translation preserves the original Russian-narration focus and those historical competitor checks. Dictator's own billing has been updated to the current shared token system.

Download Dictator to try narration with inline directions in the application. You can listen to standard voices online free and without registration.

How to recognise expressive AI speech: 6 things to check

A basic voice reads steadily, inferring its intonation from punctuation, sometimes very successfully. More deliberate delivery comes from being able to specify your intention. Check these capabilities before paying.

  1. Directions inside the script. The voice recognises instructions in square brackets, performs them, and does not speak the instruction aloud.
  2. Directions in the script's language. The original comparison focused on Russian. ElevenLabs and Google documented English tags; the documentation reviewed in October did not establish how those tags behaved inside Russian text.
  3. Vocal and background sounds. Sighing, laughter, crying or yawning in the voice itself, and audience or crowd laughter behind it.
  4. Pauses and emphasis on one word. A pause at a meaningful point and stress on an individual word rather than the whole sentence.
  5. A voice for your task. A ready-made library voice, a voice described in words, such as "low and husky, like a storyteller by a campfire," or your own voice.
  6. Access from your region. The original Russian comparison also checked whether services could be used from Russia and paid for in rubles.

AI text-to-speech services compared

Competitor allowances and regional-access statements in this table are the October 14-16, 2026 observations, rather than a fresh November survey.

ServiceHow to direct the deliveryFree optionAccess from Russia in the original checkBest suited to
Dictator Premium voicesA catalogue of 73 Russian tags for emotion, delivery, sounds, background and pauses; written descriptions in brackets; emphasis on a wordStandard Microsoft voices without a character cap or emotion tags; Premium cloud voices use Free's shared 1,000 monthly tokensYes; Windows and macOS application, with ruble pricing availableRussian videos, lessons, fairy tales and scenes with directed delivery
ERA2 VoiceEmotion through voice selection: joy, sadness, irony or whisper; stress marked with a plus, pauses with three hyphens; no tag list found300 characters after registration; then from RUB 490/month for 8,000 charactersYes; website, ruble paymentsSteady Russian narration with controlled word stress
ElevenLabsEnglish tags such as [whispers], [laughs], [sighs]; the reviewed documentation did not establish behaviour in Russian scripts10,000 credits/month, approximately 10,000 characters, without a commercial licenceRestrictedEnglish narration
ZvukogramVoice styles such as kind or angry, speed and pitch, plus SSML pauses; no sounds such as laughter found1,000 characters without registration and 2,000 after registration; then pay as you go, one token = RUB 1Yes; Russian bank cards and SBPLong texts and a choice of more than 140 Russian voices
SpeakmiRussian tags such as [радостно], [шёпотом], [пауза]; no complete list on the site30 seconds after registration; then RUB 10-13 per minuteYes; website and Telegram bot, ruble paymentsShort narration from a phone
Yandex SpeechKitNo emotion or sound tags in the reviewed markup; pauses and word stresses through markup, whispering through a marina voice settingNo new-customer grant foundYes, through Yandex CloudVoice bots and applications
SaluteSpeechSSML modes whisper, sad, annoyed and happy for Russian, plus pauses and stressNew-customer onboarding was closed from July 15, 2026Closed to new customersExisting customers
Google Gemini-TTSEnglish preview tags such as [sigh], [laughing], [whispering], and natural-language delivery descriptionsNo free allowance foundNot stated on the page; payment requiredApplication developers

In the reviewed sources, Russian inline bracket tags were described by Dictator's Premium voices and Speakmi. Dictator provided a ready-made catalogue explaining placement, background sounds, and a desktop interface. Cloud narration now spends the account's shared monthly tokens; it is no longer sold as the separate RUB 6.5/minute package described in the old Russian edition. Download Dictator.

What a script with directions sounds like

Here is the original Russian example, with a different mood for each phrase:

[радостно] Привет! Сегодня отличный день. [смеётся] Ха-ха, правда здорово! [шёпотом] Только никому не говори. [вздыхает] Эх, ладно, пора за работу.

It means: happily, "Hello! Today is a great day"; laughing, "Ha-ha, that's really wonderful"; whispering, "Just don't tell anyone"; sighing, "Oh well, time to get to work."

Listen to the original Russian sample: a recording from Dictator's tag panel, also available through its preview action · MP3

The voice sounds happy, laughs, whispers and sighs without speaking the bracketed directions. The same technique can be used in a lesson or a video:

[уверенно] Сегодня разберём одну ошибку, которую делают почти все. [пауза] Вы пишете текст для голоса так же, как для глаз. [смеётся] Ха, а голос мыслей не читает, он читает знаки препинания. [с нажимом] Запятые решают всё. [смех в зале]

In English: confidently, "Today we'll look at a mistake almost everyone makes"; pause; "You write for the voice just as you write for the eye"; laughter; "But a voice doesn't read minds, it reads punctuation"; with emphasis, "Commas make all the difference"; audience laughter. The last sound plays behind the speaker, like a studio audience.

What can you put in square brackets?

Dictator's "Tags: intonation and sounds" panel sits above the text field. The catalogue described in the original article has 73 tags in six groups. Click a tag to copy it, then paste it into the script. The Russian tags below are intentional examples from that catalogue, with English explanations.

Adjust an emotion with words meaning "slightly," "very," or "extremely": [слегка грустно] (slightly sad), [очень радостно] (very happy). When a ready-made tag is not enough, describe the delivery in your own words: [устало, почти шёпотом] (tired, almost whispering). For the complete list and placement rules, see speech synthesis: what it is and which programs to use.

Dictator's tag panel: delivery styles and sounds, with a tooltip explaining each tag
Dictator's tag panel: delivery styles and sounds, with a tooltip explaining each tag

Choose a library voice, describe one, or use your own

Directions work with Premium voices; choose the voice itself for your task.

Dictator's Create a voice from a description window: a prompt field, examples and candidates to preview
Dictator's Create a voice from a description window: a prompt field, examples and candidates to preview

How to create expressive AI narration in Dictator

  1. Download Dictator for Windows or macOS and sign in. An account is free. The original application flow offered Yandex, Google, VK, Apple and email sign-in; the available choices depend on your region.
  2. Open "Text to speech," choose Premium voices, and paste your script.
  3. Expand "Tags: intonation and sounds" above the text field. Click the tag you need and insert it into the text.
  4. Generate the audio, listen, then save MP3, WAV or OGG.
Dictator's Text to speech section: the tag panel above the script, with Premium voices and model selection on the right
Dictator's Text to speech section: the tag panel above the script, with Premium voices and model selection on the right

The original comparison describes two Premium models, available across the plans that can use Premium cloud speech. Premium 2 Pro is the default, with steady delivery and tag interpretation. Premium 3 Drama gives a more theatrical performance. You can describe the delivery in more detail, such as [тепло, как старому другу], meaning "warmly, as to an old friend," or apply a whisper or emphasis to a few words within a sentence. The reading varies between generations, so it can be worth generating two or three versions and keeping the one that works best.

Where AI narration needs help

How much does expressive narration cost?

Dictator cloud features use one shared monthly token balance, verified against the application's pricing specification on November 16, 2026:

Premium cloud speech, translation and other cloud features spend that same balance. The available narration duration depends on how much of the balance you spend on speech and on the displayed operation estimate. Paid plans can top up their tokens; top-ups are valid for 180 days. Separate voice-minute packages are no longer sold. Existing previously purchased minute balances retain their own expiry and are used according to the account's billing rules.

Microsoft standard voices and local offline voices remain free without a character cap, but do not perform Premium emotion tags. The original October catalogue counted 322 Microsoft voices in 75 languages and four Russian offline voices. These counts are dated catalogue observations, not a guarantee that the catalogue will never change.

Which AI tool fits your task?

Find your task below. Competitor recommendations refer to the capabilities reviewed in October.

Frequently asked questions

Which AI reads text like a real person?

Look for a voice you can direct. For Russian, Dictator's Premium voices offer a catalogue of 73 tags for emotion, delivery, sounds, background and pauses, and you can describe complex delivery in words. Speakmi also documented Russian tags, without a complete catalogue, in our October comparison. ElevenLabs is another option for English narration, with regional restrictions to check before using it.

How do I add emotion to AI speech?

Put a direction in square brackets at the start of the sentence. The Russian examples [радостно], [грустно] and [тревожно] mean happy, sad and anxious. The voice performs the direction until the full stop rather than speaking the tag. Add a modifier such as "very" or "slightly" to adjust its strength.

How do I insert a pause?

Dictator's Russian catalogue includes [пауза] and [долгая пауза], pause and long pause. The original tests found pauses especially pronounced with Premium 3 Drama. Punctuation also creates pauses in any synthesis engine; a full stop normally produces a longer one than a comma.

Can I generate AI speech for free?

Yes. Dictator's standard and offline voices are free without a character cap. Premium cloud speech can use the Free plan's shared 1,000 monthly tokens; those same tokens also cover other eligible cloud features. The separate one-minute trial described in the old article is not an additional allowance on top of the current token balance.

Do ElevenLabs tags work with Russian?

The October documentation reviewed for this article used English tags such as [whispers], [laughs] and [sighs] and did not specify their behaviour inside Russian scripts. The service's September 24, 2026 help page also described restrictions on access from Russia. Check its current documentation and availability before choosing it for that use.

Do I need the internet?

Premium cloud voices do: generation takes place on a server. Dictator also offers offline voices that run on your computer, but those do not perform Premium emotion tags.

Expressive delivery is easier to hear than to describe. Download Dictator and try the original tagged sample, then read the same text without tags online to compare the delivery.

Where the service information came from

Competitor terms were checked on the services' own pages October 14-16, 2026: ElevenLabs pricing and restricted-country help; ERA2 Voice pricing; Zvukogram; Speakmi; Yandex SpeechKit markup and pricing; SaluteSpeech individual pricing; and Google Gemini-TTS documentation.

Dictator's shared monthly budgets, international subscription prices and paid-plan token top-ups were checked against its pricing specification on November 16, 2026. The original audio sample and competitor research retain their original October dates.

Who wrote this

BOZEX develops Dictator (dictator.my), an application for text to speech, voice typing, transcription and translation. The Premium voice capabilities described here come from the application. Competitor terms were checked on their websites and in their documentation October 14-16, 2026; Dictator's shared token plans were checked November 16, 2026.