← Back to Projects

AI-Powered Multilingual Content Production Platform

AI • Language Technology • Media Automation • Education

An end-to-end platform that transforms text, audio recordings, and YouTube media into structured bilingual lesson packages through automated transcription, segmentation, translation, voice synthesis, audio processing, and cloud deployment.

Whisper Google Translate Google TTS Java Angular MySQL AWS OpenNLP yt-dlp ffmpeg Full-Text Search gRPC
  • One-Click PipelinesFrom source media to a deployable multilingual lesson package
  • Multimodal InputText documents, audio recordings, CDs, DVDs, and YouTube media
  • Reusable ContentSearchable multilingual segments reused across lessons
  • Cloud DeliveryAWS-hosted processing, storage, packaging, and deployment

From Raw Media to Production-Ready Lessons

The platform turns heterogeneous input sources into standardized multilingual lesson content. Each workflow is composed of reusable processing stages that can be configured according to the source type and required output. A workflow selects only the applicable stages; not every source passes through every possible operation.

Configurable twelve-stage pipeline from text, audio, discs, recordings, and YouTube media to multilingual lesson outputs
Source-aware workflow orchestration selects ingestion, recognition, segmentation, translation, voice, media, assembly, validation, packaging, and deployment stages as required.

From a single YouTube URL or source file to a production-ready multilingual lesson package with one click.

Overview

A Production Platform for Multilingual Learning Content

The platform was designed to automate the creation of high-quality multilingual educational media for language schools and individual learners. It processes raw text, recorded speech, and online video sources and converts them into reusable bilingual content.

Its responsibilities extend beyond translation. The platform orchestrates speech recognition, natural-language segmentation, machine translation, text-to-speech generation, audio extraction, timing recalculation, audio normalization, subtitle generation, lesson assembly, packaging, storage, and production deployment.

The result is a repeatable content-production process that replaces a large amount of manual transcription, translation, recording, editing, synchronization, and file preparation.

  • Automated Content Production

    Configurable pipelines turn varied source materials into consistent lesson outputs.

  • AI Service Orchestration

    Speech recognition, translation, and voice synthesis work as coordinated processing stages.

  • Reusable Multilingual Data

    Persistent text, audio, and timing records support search and cross-lesson reuse.

  • Production-Ready Lesson Packaging

    Validated lesson files and linked media assets are assembled for deployment.

System Design

Platform Architecture

A layered Java-based platform connects the Angular client, processing services, multilingual database, AWS infrastructure, and external AI providers through a configurable plugin-oriented architecture. The platform includes access-management functionality supporting free and paid subscription models.

Layered LTA architecture from Angular client and Java services through MySQL, AWS storage, processing providers, and an external player
The production platform owns ingestion, orchestration, content storage, packaging, and deployment. Lesson playback occurs in an independently developed external PWA outside this project’s scope.

Content Architecture

Lessons Built from Reusable Multilingual Segments

A segment stores text and associated audio in a specific language, while corresponding variants can represent the same content in other languages. A production lesson selects two languages and references an ordered set of hundreds of segment groups rather than requiring every media file to be duplicated.

When segments originate from audio, their records can include timing and synchronization data. Producers use filters and full-text search to find existing material and reuse it in new lessons. Lesson definitions and generated files are stored separately, with persistent metadata linking reusable records to media objects while leaving room for implementation-specific storage choices.

Lesson references reusable multilingual segment groups linked to MySQL metadata, S3 media, full-text search, and multiple lessons
The content model separates ordered lesson definitions, reusable multilingual segment variants, persistent metadata, search indexes, and linked media storage.

Processing Capabilities

A Modular AI and Media Processing Toolkit

Speech Recognition

Whisper-based transcription converts native-voice recordings into timestamped text for later segmentation and translation.

Natural-Language Segmentation

OpenNLP divides long texts into natural, reusable learning segments while preserving sentence boundaries and linguistic structure.

Machine Translation

Google Translate and configurable external services generate target-language text that can be reviewed, edited, stored, and reused.

Voice Synthesis

Google TTS and additional providers generate audio using selectable voices, referred to inside the platform as actors.

YouTube Ingestion

A yt-dlp-based module retrieves source media and extracts the required audio streams from supported YouTube content.

Audio Processing

ffmpeg converts formats, extracts channels, trims audio, recalculates timing, normalizes files, inserts pauses, and combines final lesson audio.

Subtitle Generation

Segment text and timing data are converted into synchronized SRT subtitle files.

Lesson Assembly

The builder combines ordered source audio, configurable pauses, target audio, text, subtitles, metadata, and output settings.

Search and Reuse

Filters and full-text search locate existing segments and lesson content for reuse in future productions.

Cloud Deployment

Completed packages and associated assets are published to production storage for consumption by downstream applications.

Content Repository

Create Once, Search, Adapt, and Reuse

The database is not merely a collection of generated files. It is a structured multilingual repository designed to preserve and reuse content across many lessons.

Producers can search existing segment text, filter by language and other available metadata, select previously generated audio variants, and assemble new lessons without recreating every component.

This approach reduces duplicate translation and voice-generation work while helping maintain consistency across a growing lesson library.

  • Multilingual Segment Library
  • Full-Text Search
  • Configurable Filters
  • Audio and Text Variants
  • Cross-Lesson Reuse
  • Persistent Timing Metadata
Reusable lesson repository relationships between segments, search, metadata, and stored media
A repository view of reusable content relationships, derived from the platform content model rather than an invented interface screenshot.

Workflow Examples

Multiple Inputs, One Consistent Production Model

Text to Bilingual Lesson

A producer uploads source-language text through the Angular web client. The platform divides it into natural segments, translates the segments, generates voice audio, builds synchronized output files, and deploys the finished lesson package.

  1. TXT or XLS upload
  2. OpenNLP segmentation
  3. Google Translate
  4. Selected TTS actor
  5. Source and target audio generation
  6. Pause insertion
  7. Lesson audio compilation
  8. SRT generation
  9. .lesson packaging
  10. Production deployment

Recorded Voice to Structured Lesson

Audio from a CD, DVD, recording, or other source is transcribed and divided into synchronized text and audio segments. The translated and synthesized variants are stored as reusable content before lesson production.

  1. Native voice recording
  2. Whisper transcription
  3. Timestamped text
  4. Text and audio segmentation
  5. Timing recalculation
  6. Translation
  7. Target-language TTS
  8. Database storage
  9. Lesson assembly
  10. Deployment

YouTube URL to Finished Lesson

A supported YouTube URL initiates the complete automated pipeline. Audio is extracted, transcribed, segmented, translated, voiced, synchronized, assembled, packaged, and prepared for production delivery.

  1. YouTube URL
  2. yt-dlp
  3. ffmpeg
  4. Whisper
  5. Segmentation
  6. Translation
  7. TTS
  8. Synchronization
  9. Lesson builder
  10. Packaging
  11. Deployment

End-to-End Automation

From YouTube URL to Finished Lesson

The complete media workflow connects retrieval and audio preparation with transcription, synchronized segmentation, translation, voice synthesis, lesson assembly, subtitle generation, project-specific packaging, and cloud deployment.

Sixteen-step YouTube-to-lesson workflow ending with AWS S3 delivery to an external player outside the project scope
A supported YouTube URL can drive the full production workflow. The final consumer is an external PWA player developed independently and outside the scope of this project.

Cloud Platform

AWS-Based Processing, Storage, and Delivery

The production platform is hosted on AWS and separates application execution, persistent business data, media storage, and auxiliary processing responsibilities.

AWS EC2

Hosts the Java server and principal platform services.

AWS S3

Stores source media, generated segment audio, subtitles, lesson packages, and production assets.

AWS Lambda

Supports selected background or auxiliary cloud-processing tasks.

MySQL

Stores lessons, segment metadata, multilingual text, processing state, access data, subscriptions, and references to media objects.

The architecture also uses supporting AWS services where required by deployment and processing workflows.

Source-Language SegmentConfigurable PauseTarget-Language SegmentConfigurable PauseNext Segment

Production Output

Structured Packages for Language-Learning Playback

Language direction is configurable. The lesson builder sequences selected source- and target-language variants with timing, pause, subtitle, and output settings, then produces a project-specific .lesson package and its associated assets.

  • Combined lesson audio
  • Individual segment audio
  • Source and target texts
  • SRT subtitle files
  • Timing and synchronization data
  • Lesson metadata
  • .lesson package
  • Production storage assets

System Boundary

Built for a Dedicated Language Learning Player

The generated lesson packages are consumed by a separate Progressive Web Application designed specifically for language learners. The player presents synchronized multilingual text and audio and provides the learner-facing playback experience.

LTA Content Production Platform.lesson package and media assetsExternal Language Learning Player

The Language Learning Player was developed independently by another developer and is outside the scope of my work on the LTA content production platform.

Technology Stack

Technologies Behind the Platform

FRONTEND

AngularDevExpress

BACKEND

JavaPlugin-oriented service integrations

DATA

MySQLFull-text searchAWS S3 object storage

AI AND LANGUAGE SERVICES

WhisperGoogle TranslateGoogle TTSAdditional TTS providersOpenNLP

MEDIA PROCESSING

yt-dlpffmpegSRT subtitle generation

CLOUD

AWS EC2AWS S3AWS LambdaSupporting AWS services

My Contribution

My Role

I designed and developed the LTA platform as an end-to-end content production system, covering product architecture, full-stack implementation, AI integrations, multimedia workflows, data modeling, cloud infrastructure, and automated deployment.

  • Product architecture and technical design
  • Angular and DevExpress web-client development
  • Java server development
  • MySQL schema and multilingual content-model design
  • Workflow orchestration and automation
  • Whisper speech-recognition integration
  • Google Translate integration
  • Google TTS and additional voice-provider integrations
  • OpenNLP segmentation integration
  • yt-dlp-based media ingestion
  • ffmpeg-based audio processing
  • Segment timing and synchronization workflows
  • Lesson builder and .lesson package generation
  • Full-text search and reusable-content functionality
  • AWS EC2, S3, and Lambda integration
  • Access and subscription management
  • Production deployment and maintenance