YouLaf: AI video dubbing into Azerbaijani, controlled by a prompt
Who: Azerbaijani-speaking creators, teachers and small media or training teams who want foreign-language videos (tutorials, lectures, product demos) available in Azerbaijani. Problem today: There are only two ways to do this. Subtitles, which viewers skip while watching. Or manual dubbing, which needs a translator, a voice actor and an editor, so it is slow and expensive and does not scale to a single tutorial. Simple machine dubbing has its own faults we hit ourselves: the new audio drifts away from the picture, and English technical terms (Xamarin, GitHub) are read with Azerbaijani letters, so they sound wrong. Our solution: YouLaf takes an MP4 or YouTube link and a plain-language request ("female voice, formal tone, dub it") and returns the same video in Azerbaijani, with each sentence spoken at its original moment. Terms keep their original spelling on screen but are pronounced correctly. A 60-second video is dubbed in under a minute on the live server.
Preliminary score
- Value for the user19.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI25.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing15.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility11.0 / 15
Data requirements, running costs and a clear next step.
- Originality6.0 / 10
A meaningful difference from common solutions.
A finalist: the live jury scored it 79.0 / 100 on stage.
What the judges wrote
- Claude Opus 5.5Preliminary75/100
- Strengths
- Your pipeline is real, deployed and has a clear AI role at every step. Whisper timestamps drive sentence-level sync, Gemini translates while protecting terms and writing phonetic pronunciations, and FFmpeg fits the result to the original timing. The Xamarin/Blazor example on slide 5 shows exactly why your term handling matters for technical content. Your testing section is unusually honest and specific: Google Translate rate limits, Gemma 3 turning "C-Sharp" into "Zamarin", Whisper's "You Laugh", the forced TTS failure to verify the fallback, and YouTube blocking in three of four runs. Your competitor comparison is candid and positions you clearly around terminology control and self-hosting.
- To improve
- Your translation and pronunciation quality is still judged only by your own team, so run a small blind evaluation with native Azerbaijani speakers. Rate 10–20 clips for accuracy, naturalness and term correctness, and compare against ElevenLabs or HeyGen output on the same clips. Your user value would be stronger if you picked one segment, such as corporate onboarding, and measured the outcome against the current approach. Examples are cost and turnaround versus a studio, or comprehension versus subtitles. On feasibility, estimate the per-minute cost once you leave the free Gemini tier, and name a first pilot customer or partner as your concrete next step.
- GPT-6 AstraPreliminary78/100
- Strengths
- Your sentence-timed dubbing pipeline addresses concrete localization problems for Azerbaijani training teams, especially technical-term pronunciation. Your screenshots, worked translation example and measured jobs support an end-to-end workflow with AI essential to transcription, translation and speech generation. Your testing report provides valuable specifics, including repeated rate-limit failures, before-and-after transcription examples, safety-limit checks and a deliberately forced voice-fallback test.
- To improve
- Your next evaluation should use varied real training clips rated by native Azerbaijani speakers, measuring translation accuracy, pronunciation and sentence-level timing error; matching total video duration alone does not establish synchronization. Run a small training-team pilot against manual dubbing or an existing tool, reporting review effort and cost per finished minute. Quantify hosting and paid-API costs beyond free quotas, and clarify which video-derived data reaches external providers even when the application is self-hosted.
The team
Ayşen Qehremanova
developer
Elshan IbrahimovLead
developer
Afraim Mammadov
developer