System Requirements Document for neongreen

1. Introduction

neongreen is the project under which LingoDub Studio is built: a professional, AI-powered video dubbing application. The product exists to take a full-length video in one language and deliver a complete, separately saved dubbed video in another language, where the original picture quality and resolution are preserved, the original background music, sound effects, ambience, and cinematic audio survive intact, and the newly generated dialogue sits on top of that preserved bed in the original timing.

The intent is deliberately end-to-end rather than sample-based. A user uploads MP4 or another common video format; the system detects the source language; the user picks a target language from 170+ supported languages; the entire video — beginning to end, not a short sample — is transcribed with accurate timestamps and speaker identification; speaker changes are detected and 20+ different speakers/characters are supported; every line of dialogue is translated naturally while preserving meaning, context, emotion, and timing; natural-sounding dubbed voices are generated across male/female voices, accents, emotions, and speaking styles; optional voice cloning from an uploaded voice sample is available where supported; dialogue is separated from background audio and then remixed with the preserved original background; the dubbed speech is synchronized to the original video's timing as closely as possible; subtitles/captions can be enabled, disabled, or edited; a voice editor lets users change speaker voices, volume, timing, emotion, pitch, and speed; progress and clear error messages are always visible; and the finished work is delivered as a separate Dubbed Video that can be previewed and saved, with working video and audible dubbed audio.

Audience: creators who need their own videos speaking in another language, localization teams working through multi-speaker content, studios and post-production operators handling emotionally nuanced multilingual video, and media operators processing long-form material. The interface is a dark production cockpit for that audience — technically advanced and cinematic, but operationally precise, because the pipeline it exposes is genuinely complex and must remain legible and controllable.

2. System Overview

LingoDub Studio is a first-party web application with custom UI and application-owned identity. It orchestrates a chain of specialized services — speech-to-text, translation, text-to-speech and voice generation, speaker detection, audio separation, and FFmpeg-based video/audio processing — and never silently drops the original video or the dubbed audio at any point in that chain.

Current delivery boundary. Six surfaces carry all current accepted behavior: Landing (anonymous public entry), Sign Up and Login (anonymous identity access), Dubbing Studio (upload through processing, behind login), Voice Editor (voice/audio editing, subtitle settings, and audio mixing, behind login), and Outputs (Original Video and Dubbed Video sections with preview and download/save, behind login). No other product surfaces are in scope for the current horizon.

Actors. Two human personas do the accepted work: the Video Dubbing Creator, who owns the upload-to-processing lifecycle, and the Voice and Audio Editor, who owns the refinement lifecycle on the produced dub — speaker voice assignment, volume, timing, emotion, pitch, speed, subtitle enable/disable/edit, and the audio mix of new dialogue against the preserved background. Supporting non-human participants are the speech-to-text, translation, text-to-speech/voice-generation, speaker-detection, audio-separation, and FFmpeg-based processing services, which are typed as system/service actors and are not personas.

Ownership. All accepted human-facing work is owned by first-party destinations. Landing, Sign Up, and Login are anonymously reachable. Dubbing Studio, Voice Editor, and Outputs require an authenticated session. Background automation performs the pipeline execution itself; every human initiation, selection, decision, edit, preview, and save is owned by a first-party page.

Exclusions. The current scope does not include subtitle export files, per-role permission models, team invitation or provisioning, billing, or a public gallery of dubbed videos. These are not current capabilities and must not appear as current pages, requirements, or acceptance criteria.

2a. Product Interpretation and Delivery Boundary

LingoDub Studio is delivered as a first-party web application with custom UI. The whole user-facing pipeline — upload, source language detection review, target language selection, speaker detection review, voice selection, optional voice cloning, dubbing with live progress and errors, voice/audio editing, subtitle settings, audio mixing, and the paired Original/Dubbed output with download — is owned by first-party pages and is not delegated to an external surface.

Access ownership. Landing, Sign Up, and Login sit outside the protected boundary. Dubbing Studio, Voice Editor, and Outputs require login, because each of them depends on durable, actor-specific project state: an uploaded video and its detected language, a dub that is still processing, a speaker cast and caption set that must survive a return visit, and a completed output that must remain retrievable and bound to the correct creator. The interaction that establishes access lives on Sign Up and Login, which are anonymously reachable; no protected page owns the interaction that grants access to itself. Self-service enrollment is used because the sources establish no invitation, provisioning, or deployment bootstrap boundary. No differentiated permissions or role-based visibility are established by the sources, so access is a single authenticated continuity boundary rather than a permission model.

Service ownership. Speech-to-text, translation, text-to-speech/voice generation, speaker detection, audio separation, and FFmpeg-based video/audio processing are reliable third-party or first-party services invoked by the application. Their internal surfaces are not product pages and are not exposed to users as destinations; users see their progress, results, and errors through first-party pages.

Current versus future. Everything in this document is current. The scalable-architecture requirement is a current architectural obligation whose results — additional AI models, additional languages beyond the current set, additional voices, and additional video-processing features — are future-facing. Those future additions must be able to arrive without rework, but no future model, language, voice, or video-processing feature is a current capability, destination, or acceptance criterion.

2c. Page Content and Component Coverage

Landing

• Purpose and information. Anonymous public entry that explains, before any account exists, that LingoDub Studio is a professional AI-powered video dubbing application for full-length videos: 170+ target languages, 20+ speakers/characters, whole-video processing, preserved original audio bed, and a separately delivered dubbed video. The page is an asymmetric split composition — oversized editorial type on the left, a large cinematic dubbing scene on the right — never centered copy above a generic dashboard mockup.
• Primary actions. Start a dub (the only conversion action; leads to Sign Up) and Watch the pipeline, a quiet outlined control beneath the headline that reveals the in-page Signal Path explainer.
• Supporting actions. Header entry to Login for a returning creator or editor; scroll navigation across the page's own sections.
• Domain entities (read-only, presentational). Pipeline stage names (Upload, Detect, Script, Translate, Cast, Mix, Deliver); language-pair example (ES → Japanese); speaker example (SPEAKER 04); mix example (MIX PRESERVED); timecode example (00:18:42).
• Component responsibilities. Hero headline block; dimensional dubbing chamber; floating glass metadata labels orbiting the chamber; Signal Path rail preview listing the seven numbered luminous stations connected by a live processing line; split-spectrum timeline illustration with pale original-dialogue waveform lanes on the left and neon-green dubbed-language voice lanes on the right joined by thin animated translation vectors; CTA pair.
• Loading state. The chamber initializes progressively; while it initializes, the composed still frame of the chamber is shown so the hero never appears empty.
• Empty state. Not applicable — this page has no user data.
• Success state. Not applicable — no data is committed here.
• Error state. If the dimensional scene cannot initialize, the page falls back to the composed static first frame with all headline, metadata, and controls intact and fully operable.
• Recovery. The fallback is permanent and non-blocking; the visitor can still reach Sign Up and Login. No error is surfaced as a product failure.

Sign Up

• Purpose and information. Anonymous self-service enrollment that establishes the application-owned identity required for durable dubbing projects and outputs. Explains, in one line, that an account is what keeps uploaded videos, processing runs, speaker casts, and completed outputs available across visits.
• Primary actions. Create account — submits the enrollment form and establishes the session.
• Supporting actions. Field-level validation feedback as the user types; a link across to Login for a returning visitor.
• Inputs. Name, email address, and password. These are the identity fields required to privately own and resume durable actor-specific state.
• Domain entities. Creator account identity; authenticated session.
• Component responsibilities. Enrollment form with inline validation; password entry with show/hide; submit control with in-flight disabled state; cross-link to