Text to Speech: PDF, DOCX → MP3 + Subtitles, Ukrainian +22 More

Turn text, articles, PDFs and Word documents into natural-sounding speech — MP3, WAV or OGG — with SRT and VTT subtitles. Neural voices run inside the Actor: no API key, no account elsewhere, no per-character fees.

Listen first: Ukrainian — Mykyta (recommended) · Ukrainian — Tetiana · English HQ — LJSpeech · English fast — John · German — Thorsten

Why this Actor

	This Actor	Typical alternatives
Commercial use	Every voice is CC0, public domain, Apache 2.0 or CC BY 4.0 — safe for YouTube, courses, ads and client work	Many free voices are "non-commercial only" (CC BY-NC); unofficial Edge or Google voices may break those services' terms
Ukrainian	3 natural high-quality Ukrainian voices (українська озвучка тексту), word stress from a Ukrainian dictionary, lively intonation	Rarely offered at all; open voices often stress Ukrainian words wrongly
Documents in	PDF, DOCX, HTML, Markdown, TXT — menus, scripts and Markdown symbols skipped	Plain text only; you copy and clean the text yourself
Subtitles	SRT + VTT timed per sentence, free	Extra tool or extra fee
Cost	Per minute of audio; no API key or subscription elsewhere	ElevenLabs-style wrappers need your own paid key and bill per character

Features

• 37 neural voices in 23 languages: Ukrainian, US and UK English, German, French, Spanish, Italian, Polish, Portuguese, Dutch, Czech, Russian, Romanian, Finnish, Hungarian, Slovak, Swedish, Danish, Greek, Latvian, Slovenian, Chinese, Persian, Vietnamese.
• Speed control 0.5×–2×, natural pauses between sentences and paragraphs.
• Delivery: "Lively" (more varied intonation) or "Standard" (even and calm); "Auto" uses Lively for Ukrainian.
• Audiobooks, podcasts, e-learning, accessibility: long texts (up to 1,000,000 characters per item) are read sentence by sentence.

Input

{
  "text": "Hello! This is a natural-sounding neural voice.",
  "sources": [{ "url": "https://example.com/article.pdf" }],
  "voice": "en_US-john-medium",
  "speed": 1.12,
  "style": "auto",
  "format": "mp3",
  "subtitles": true
}

Field	Meaning
text	Text to read; blank lines separate paragraphs
sources	Links to TXT, Markdown, HTML, PDF or DOCX files — one audio file each. Files in your own Apify key-value store work too: set the store's access to "Anyone with ID can read" and use https://api.apify.com/v2/key-value-stores/<store ID>/records/<key> (the Actor runs with limited permissions and does not use your token).
voice	Voice ID from the table below
speed	0.5–2; default 1.12 (smoother, fewer hesitations than the voice's own pace 1.0)
style	auto (default), lively or standard
format	mp3 (64 kbit/s), wav, ogg (Opus)
subtitles	Also return .srt and .vtt
maxCharactersPerItem	Longer texts are cut; ~750–900 characters ≈ 1 minute of speech

Voices

Language	Voice IDs
Ukrainian	uk_UA-tetiana-high, uk_UA-mykyta-high, uk_UA-oleksa-high
English (US)	en_US-ljspeech-high, en_US-john-medium, en_US-kristin-medium, en_US-bryce-medium, en_US-norman-medium, en_US-joe-medium
English (UK)	en_GB-cori-high, en_GB-alba-medium
German	de_DE-thorsten-high, de_DE-kerstin-low
French	fr_FR-siwis-medium, fr_FR-gilles-low
Spanish (Spain)	es_ES-davefx-medium
Spanish (Mexico)	es_MX-claude-high
Italian	it_IT-serena-high
Polish	pl_PL-gosia-medium, pl_PL-darkman-medium
Portuguese (Brazil)	pt_BR-faber-medium
Dutch	nl_NL-pim-medium
Czech	cs_CZ-jirka-medium
Russian	ru_RU-denis-medium, ru_RU-dmitri-medium
Romanian	ro_RO-mihai-medium
Finnish	fi_FI-harri-medium
Hungarian	hu_HU-anna-medium
Slovak	sk_SK-lili-medium
Swedish	sv_SE-nst-medium
Danish	da_DK-talesyntese-medium
Greek	el_GR-rapunzelina-medium
Latvian	lv_LV-aivars-medium
Slovenian	sl_SI-artur-medium
Chinese	zh_CN-chaowen-medium
Persian	fa_IR-amir-medium
Vietnamese	vi_VN-vais1000-medium

high voices sound best and cost more per minute; medium and low are a little more synthetic, but about 7× faster.

Output

{
  "url": "https://example.com/article.pdf",
  "voice": "en_US-ljspeech-high",
  "language": "English (US)",
  "format": "mp3",
  "characters": 5240,
  "truncated": false,
  "durationSeconds": 352.4,
  "sentences": 61,
  "audioUrl": "https://api.apify.com/v2/key-value-stores/.../records/0002-article.mp3",
  "srtUrl": "https://api.apify.com/v2/key-value-stores/.../records/0002-article.srt",
  "vttUrl": "https://api.apify.com/v2/key-value-stores/.../records/0002-article.vtt"
}

A document that cannot be downloaded or read gets an error field instead and is not charged.

Pricing

Event	Price	What counts
audio-minute	$10 per 1,000 (1 cent a minute, $0.60 per hour)	one started minute of audio with a medium or low voice
hq-audio-minute	$40 per 1,000 (4 cents a minute, $2.40 per hour)	one started minute of audio with a high voice (needs ~7× more computing)

Plus Apify's standard start fee of $0.05 per 1,000 runs. Subtitles are free. The run stops before it would exceed your maximum charge per run.

Related Actors

• Audio & Video Transcriber — the other direction: speech to text and subtitles
• Document to Markdown for AI — PDFs, scans and Office files → Markdown
• Image to Text OCR — scanned pages to text, then to speech

Open source: github.com/AndriyVK/-apify-document-tools

Use with AI agents and code

• MCP (Claude, Cursor, VS Code, any MCP client): add https://mcp.apify.com?tools=ai_admin/text-to-speech as an MCP server — the agent gets one tool with this Actor's input schema and pays per use through your Apify account.
• One HTTP call that returns the results:

curl -X POST "https://api.apify.com/v2/acts/ai_admin~text-to-speech/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello! This is a natural-sounding neural voice.", "voice": "en_US-john-medium", "format": "mp3", "subtitles": true}'

Limits

• Scanned PDFs without a text layer are not read — run them through Image to Text OCR first.
• Numbers, abbreviations and foreign words are read by the voice's rules and may sound unusual.
• Speech engine: Piper (GPL-3.0). Voice models: rhasspy/piper-voices; CC BY 4.0 voices (en_GB-alba, fr_FR-siwis, it_IT-serena, sl_SI-artur, vi_VN-vais1000) are credited to their dataset authors as listed in each voice's model card.