Friendly Reader
Accessible reading for everyone.
Friendly Reader is a free web app that turns any PDF — a school book, a care guide, scanned notes, a photocopied chapter, an out-of-print novel, a technical document — into an accessible, audible, easier-to-understand reading experience in 70+ languages.
It brings together three advanced technologies in service of accessibility: ElevenLabs v3 for natural speech with prosody, emotion, and multi-voice dialogue; Gemini 2.5 for plain-language rewriting, OCR, and page classification; and an accessibility-first interface built around large type, high contrast, keyboard navigation, and screen-reader support.
Many important texts do not exist in accessible formats.
They are not on Audible, Storytel, Spotify, or any commercial platform. Friendly Reader fills this gap, and does so for free for the people who need it.
Anyone who's heard "this book is not available."
From upload to a page being read aloud, in six steps.
- 01
Sign in with an access code
No email, no password, no social login.
- 02
Upload a PDF
From desktop, phone, or tablet. Text PDFs are extracted directly; scanned PDFs are run through Gemini OCR.
- 03
AI classifies the pages
Identifies real content vs cover pages, TOC, copyright, dedications, blanks — so reading starts where the book actually begins.
- 04
Read page by page
Large typography, high contrast, simple navigation. Keyboard shortcuts for everything.
- 05
Press Listen
The page is synthesized with ElevenLabs v3. Dialogue can be rendered with distinct voices, each character keeping consistent voice across the entire book.
- 06
Toggle plain text
Each page can be rewritten by Gemini in clearer language — short sentences, everyday words, same meaning. Audio can then be generated from the simplified version.
Everything runs in the browser. No installation, no ads, no tracking, no cost.
Not a TTS engine. A performance of the text.
Human speech, not machine reading
Traditional text-to-speech becomes tiring after minutes: flat prosody, questions sound like statements, exclamation marks have no force, dialogue loses meaning. ElevenLabs v3 captures the rhythm of punctuation, the rise of a question, the pause of a comma, the hesitation of an ellipsis. The result sounds like a person reading aloud.
Emotion tags
When the text makes emotion clear — a whisper, a shout, sadness, curiosity, laughter — Friendly Reader can prepare the audio with emotion tags that ElevenLabs v3 understands:
[whispering] [shouting] [curious] [sad] [laughs] [crying]
This matters in literature, children's books, theater, testimony, memoir, educational material. Friendly Reader does not merely read; it performs.
Automatic multi-voice dialogue
When a page contains dialogue, Gemini 2.5 Pro analyzes it and divides it into speaking turns — narrator, named characters, and unnamed voices. Each speaker is assigned a distinct voice from a curated ElevenLabs catalog, and that voice persists across the entire book: if "María" appears on page 12 and again on page 87, she keeps the same assigned voice. The page is then synthesized using ElevenLabs' text_to_dialogue as a natural conversation. Close to a professional audiobook experience generated from an ordinary PDF, without a studio.
Plain-language rewriting
Audio makes the format accessible. But sometimes the barrier is not seeing the text — it's understanding it: dense academic prose, old writing, technical vocabulary, legal language. When "Plain Text" is enabled, each page is rewritten by Gemini preserving meaning while replacing complex words with everyday alternatives, in the original language. The simplified version can be read or listened to, interchangeably with the original.
Truly multi-language
ElevenLabs v3 supports 70+ languages — Spanish and regional variants, English, Portuguese, French, Italian, German, Catalan, Galician, Basque, Romanian, Polish, Arabic, Hebrew, Hindi, Japanese, Korean, Mandarin, Indonesian, Vietnamese, Filipino, Swahili, and many more. Voices that pronounce the source language naturally, not transliterations.
Accessibility-first interface
- Large, comfortable typography and high contrast
- Strong visible focus outlines and keyboard navigation
- Large touch targets for tremors and imprecise pointing
- aria-live regions and role=alert for screen-reader updates
- Correct language metadata so speech engines pronounce content right
- No aggressive animations, pop-ups, or distracting patterns
For every book read, we pay so the user doesn't.
Friendly Reader is free for the user because the Escrich Foundation directly provides the technologies that make it work: high-quality voice, AI, storage, compute. The user does not pay a subscription, buy credits, see ads, or exchange personal data for access.
Reference: a 100-page book at ~2,000 chars/page, full use with multi-voice dialogue and plain-language rewriting enabled.
| Provided by the Foundation | Per 100-page book |
|---|---|
| ElevenLabs v3 — emotional speech synthesis, including dialogue | $30 – $60 |
| Gemini 2.5 Pro — dialogue analysis and speaker turns | $0.60 – $1.30 |
| Gemini 2.5 Pro — plain-language rewriting | $1.10 – $2.10 |
| Gemini 2.5 Pro / Flash — page classification | $0.05 – $0.15 |
| Gemini OCR — only when the PDF is scanned | $0 – $1.50 |
| Storage, bandwidth, and compute | $0.20 – $0.50 |
| Total Foundation contribution per book | ~$32 – $65 |
In practice, with aggressive caching, partial reading, and optional plain text, the realistic average drops to ~$10–$20 per book.
For roughly $7.50 per person per month, the Foundation provides a service that costs the user $0 — no ads, no data sale, no subscriptions. A private equivalent (audiobook + speech-reading + plain-language tools) would cost the user over $300/year, and most of the texts they need still wouldn't exist on those platforms.
The combination is what's unique.
Many tools read text aloud. Many platforms sell audiobooks. A few apps simplify text. Friendly Reader doesn't compete with any single one of them — it covers a gap none of them fills completely.
| Capability | Friendly Reader | Audible / Storytel | Speechify | Kindle TTS | Accessible libraries |
|---|---|---|---|---|---|
| Upload your own PDF | Yes | No | Yes | Yes | Rarely |
| Audiobook-quality speech | Yes (ElevenLabs v3) | Yes (human) | Sometimes | Limited | When available |
| Multi-voice dialogue | Yes, automatic | Only if recorded | No | No | Rarely |
| Emotion-aware reading | Yes | Depends | Limited | No | Depends |
| OCR for scanned PDFs | Yes (Gemini) | No | Limited | Limited | No |
| Plain-language rewriting | Yes, per page | No | No | No | Rare |
| 70+ language support | Yes | Catalog-limited | Limited | Limited | Catalog-limited |
| No ads, tracking, data sale | Yes | No | Usually no | Depends | Usually yes |
| Cost to user | Free | Subscription | Subscription | Free, limited | Eligibility required |
Built within the law, not around it.
Friendly Reader is not a public library, not a replacement for Audible, not a file-sharing service. It hosts no catalog. It only processes files each user uploads into their own private space. The service is designed around international accessibility exceptions: the Marrakesh Treaty (WIPO, 2013), the Chafee Amendment (17 U.S.C. § 121) and Marrakesh Treaty Implementation Act of 2018 in the United States, EU Directive 2017/1564, and equivalent national frameworks.
- No catalog, no public library — only what each user uploads privately
- Strict user isolation — one user cannot see or search another's files
- Mandatory attestation: users confirm their right to use the file under the applicable accessibility exception
- Acceptance logged by version, date, IP, and user agent
- No redistribution: PDFs, extracted text, and audio are not shared, indexed, or published
- No model training by the Foundation on uploaded content
- Public DMCA / Marrakesh takedown process for rights holders
- Deletion on request; access revocable on misuse
- Versioned legal documents (Terms, Privacy, Copyright, Takedown) in English and Spanish
The technology that makes it possible.
| Frontend | React + Vite + TypeScript |
| Backend | FastAPI (Python) |
| Data | SQLite + SQLAlchemy; scalable options for production |
| PDF processing | pypdf + Gemini OCR |
| AI | Gemini 2.5 Pro / Flash / Flash-Lite |
| Voice | ElevenLabs v3 — text_to_speech and text_to_dialogue |
| Storage | Local in dev; cloud object storage in production |
| Deployment | Cloud Run / container-based |
| Audio cache | Keyed by book, page, variant, voice — generated once |
Already shipping. More on the way.
- PDF upload, OCR, and page classification
- Page-by-page accessible reading interface
- ElevenLabs v3 single-voice synthesis
- Multi-voice dialogue synthesis with text_to_dialogue
- Emotion-aware audio tags
- Curated voice catalog with persistent characters per book
- Plain-language simplification with Gemini
- Continuous listening / auto-advance between pages
- Access-code login and private user spaces
- Legal framework: Terms, Privacy, Copyright, takedown procedure
- Production deployment on cloud infrastructure
- Bookmarks and notes by page
- Adjustable reading speed
- Export full book as audiobook file
- EPUB support in addition to PDF
- Offline-first mode after initial generation
- More interface languages
- Integrations with schools, libraries, and accessibility programs
Free · Accessible · Patient-first.
If this touches your life — or someone you love — it matters. Request an access code, or help us reach more people.