AI solution for translation

Technical assignment for the development of a service for automatic translation of YouTube videos from English to Russian using AI

1. General information

Goal: Development of a service that automatically downloads videos from YouTube, recognizes and translates speech from English to Russian, creating subtitles or voiceovers.

Main requirements: High processing speed, translation accuracy, process automation.

2. Functional requirements

2.1. Main functions

1. Video upload

• User input of the YouTube video link

• Ability to upload video files from PC

• Duration limit for videos (optional)

2. Speech recognition (Speech-to-Text)

• Automatic extraction of the audio track from the video

• Recognition of English speech using ASR models (e.g., OpenAI Whisper, Deepgram, Vosk)

• Splitting text into timestamps (timecode)

3. Translation (Machine Translation)

• Translation of text from English to Russian

• Use of neural network models (e.g., DeepL API, OpenAI GPT, Google Translate API)

• Preservation of timestamps during translation

4. Subtitle generation

• Formation of subtitle files (.srt, .vtt)

• Ability to edit the translation before final export

5. Voiceover of the translation (Text-to-Speech, TTS)

• Generation of voiceover for the translated text synchronized with the video

• Use of TTS engines (e.g., ElevenLabs, Microsoft Azure TTS, Google WaveNet)

• Adjustment of voice speed and intonation

6. Export of results

• Saving the translated video with the overlay voiceover

• Downloading subtitles separately (.srt, .vtt)

3. Non-functional requirements

3.1. Performance

• Processing a 10-minute video in no more than 5-10 minutes

• Speech recognition accuracy of at least 85-90%

• Translation correctness of at least 90% (when using neural network models)

3.2. Technology stack

• Backend: Python (FastAPI, Django)

• ASR: OpenAI Whisper / Deepgram / Vosk

• Machine Translation: OpenAI GPT-4 / DeepL API / Google Translate API

• TTS: ElevenLabs / Microsoft Azure TTS / Google WaveNet

• Video processing: FFmpeg

• Database: PostgreSQL / MongoDB

• Frontend: React / Vue.js

3.3. Integrations

• YouTube API (for video uploads)

• Cloud services for AI models (OpenAI, DeepL, Google Cloud)

4. Interface requirements

1. Ease of use – intuitive interface with minimal steps for the user.

2. Personal account – history of processed videos, ability to re-download.

3. Settings – choice of translation quality (basic / advanced), choice of voice for voiceover.

Expected result: A working service with high translation accuracy, fast processing, and a user-friendly interface.

Send your proposal with price, deadlines, and cases you have done in the AI niche in private messages immediately.

Do not send overpriced solutions.