Subtitles That Sound Like Real Spoken Dialogue
Bridging the gap between rigid textbook Sinhala and natural everyday spoken conversation for Sri Lankan movie fans, subtitlers, and creators.
Who Runs SriSubs
SriSubs is an independent, creator-driven web project published and maintained by Akash Makes. The project was built out of frustration with existing automated translation tools that produce awkward, unnatural Sinhala subtitles for cinema and television dialogue.
Akash Makes
Developer & editor, SriSubsAkash Makes is an IT undergraduate at SLIIT and a full-stack developer, with hands-on experience building web applications in React, Next.js, and Node.js. He built SriSubs to help Sri Lankans get Sinhala subtitles that actually sound like natural spoken Sinhala, not stiff textbook translations. He maintains SriSubs independently, handling everything from the translation engine to the content on this site.
Why SriSubs Exists: Spoken vs. Written Sinhala
The Sinhala language exhibits a classic linguistic phenomenon known as diglossia — a dramatic structural split between formal written Sinhala (ලිඛිත සිංහල) and conversational spoken Sinhala (කථික සිංහල).
Written Sinhala utilizes complex literary verb inflections, distinct grammatical pronouns, and archaic syntactic structures. While written Sinhala is revered for literature, official government documents, and scholarly journals, it sounds jarringly unnatural when spoken by characters in modern films, television series, or online videos. Hearing an action-movie hero or comedy character deliver dialogue in textbook written grammar pulls viewers completely out of the story.
Before & After Examples:
“ඔබ මේ මොහොතේ කොහේද යන්නේ?”
Sounds like a grammar workbook recitation.
“දැන් කොහෙද ඔය යන්නේ?”
Authentic phrasing heard in daily Sri Lankan life.
“ඔහු එම ස්ථානයෙන් පිටව ගියේය.”
Archaic literary past-tense verb ending.
“එයා කලින්ම එතනින් ගියා.”
Casual, natural, and emotionally resonant.
How the Translation Pipeline Works
SriSubs was engineered to maintain privacy, speed, and accuracy by splitting the subtitling workflow into distinct, transparent stages:
1. Local Client-Side Subtitle Parsing
When you drag and drop an English .srt or .vtt file into SriSubs, the file structure is parsed directly within your browser’s JavaScript engine. Timecodes, cue indices, and line breaks are cataloged in browser memory. Video media is never uploaded.
2. Cue Batching & Google Gemini AI Processing
Dialogue lines are packaged into structured batches (typically 12 lines per batch) and sent over TLS-encrypted connections to our backend endpoint. The request is processed by Google Gemini AI models guided by detailed system instructions tuned for conversational Sri Lankan Sinhala, contraction handling, and dialogue naturalism.
3. Automated Quality & Script Verification
Our backend automatically verifies that output arrays match the input line count, flags any inadvertent Latin characters or fused words, and runs targeted retries or phonetic transliteration when needed to ensure clean Sinhala Unicode script.
4. Interactive Human Review Step
Rather than treating AI output as final gospel, SriSubs renders English and Sinhala cues in an interactive, side-by-side preview table. Subtitlers can search, filter, inline-edit any line, and correct nuances before generating the final file.
Known Limitations of AI Subtitle Translation
While modern large language models represent a quantum leap over legacy dictionary-based translation, automated subtitle generation has recognized limitations that every subtitler should understand:
- Cultural Puns and Slang: Hyper-local jokes, wordplay, and contemporary regional slang can occasionally be misunderstood or translated too literally.
- Contextual Ambiguity: When a single short subtitle cue contains only one or two words (e.g., “Right” or “Shoot”), the AI may lack surrounding visual context to determine whether it signifies agreement, direction, firearms, or cinematography.
- Subtitle Reading Speed (CPS): Sinhala words are phonetically rich and often take more characters than concise English phrases. Subtitlers must occasionally trim verbose translations to respect comfortable reading speeds (15–20 characters per second).
- Nuanced Honorifics: Sinhala has multiple levels of politeness (e.g., formal vs. casual forms of address). While our prompt biases toward natural conversational intimacy, specific scenes depicting courtrooms or strict hierarchies may require human adjustment.
Editorial & Correction Policy
We are dedicated to maintaining the highest linguistic and technical quality across SriSubs. Because language evolves and AI models can stumble on edge cases, user feedback is vital to our development process.
How We Handle Reports:
If you discover a translation defect, broken character rendering, a defect in our player configuration guides, or an inaccurate prompt outcome, please contact us at akashdealwis3@gmail.com. Please include the English line, the unexpected Sinhala output, and your suggested correction.
We review reports and fix errors. While we are a small independent team and cannot offer live 24/7 phone support, incoming feedback directly informs updates to our system prompts, transliteration dictionaries, and documentation.
Privacy & Data Security Summary
Your subtitle files belong strictly to you. Subtitles are parsed locally inside your browser runtime and translated in transient server memory per request. Your subtitle files and transcripts are not stored on our servers or in any database.
For users who sign in via Google OAuth, we manage session tokens securely. Personal Gemini API keys supplied in settings are encrypted at rest with industry-standard AES-256-GCM encryption. For comprehensive information on data practices, cookies, and Google AdSense advertising, read our full Privacy Policy.
Contact & Inquiries
Have questions about SriSubs, feedback on spoken translations, or copyright inquiries? Visit our Contact Page or reach out directly by email: