BAK-Voice
AI video dubbing that keeps emotion and identity
Start new thread
Kaushal Agarwal

BAK-Voice - AI video dubbing that keeps emotion and identity

by•
BAK-Voice is an AI-powered video dubbing platform that translates content while preserving the original speaker’s voice, tone, and emotion. Upload a video + subtitles, pick a language, and export a dubbed version with voice cloning (5–23 languages), smart subtitle translation, audio enhancement, and music separation. Built for creators and educators to localize videos in minutes - not weeks, without a studio. New users get 3 free jobs to try it.

Add a comment

Replies

Best
Kaushal Agarwal
Maker
📌
🚀 From Personal Frustration to an AI-Powered Product: Launching BAK-Voice 👉 https://bakvoice.com Five years ago, while watching a Turkish series with English subtitles, I realized something: I was spending more time reading than watching 😕 That frustration sparked a question that stayed with me: What if videos could be translated into any language - without losing the original speaker’s voice, emotion, and identity ⁉️ Today, I’m excited to share BAK-Voice, a solo-built project born from a personal idea: An AI-powered video dubbing platform that preserves the original voice while translating content across languages. 🎁 New users get 3 free video processing jobs to try BAK-Voice. 🎯 What BAK-Voice Does Upload a video & subtitle file → choose a language → get high-quality dubbed output in minutes. Key capabilities: ✅ AI voice cloning across 5–23 languages (English, Chinese, Japanese, Korean, Cantonese) ✅ Intelligent subtitle translation & editing (auto language detection) ✅ AI-based audio enhancement ✅ Background music removal & subtitle refinement ✅ Secure, AWS-powered infrastructure with encrypted storage 📊 Early Validation (Before the Product) I tested the idea by dubbing anime trailers on YouTube - no marketing, just experimentation: • 195K+ organic views • 1.3K+ watch hours • 195+ subscribers That traction convinced me this problem was worth building for. ⚙️ Tech Stack (Built Solo) • Frontend: React + TypeScript + Material UI • Backend: AWS Lambda (Python), API Gateway, DynamoDB • AI Pipeline: TTS models, voice enhancement, LLM-based translation • Infra: CDK, S3, CloudFront, SQS, EC2 workers • Security: Cognito, WAF, encrypted storage 🛠️ Current Status: MVP (Expect Rough Edges) ⏰ Processing available 12 PM – 12 AM PT (cost-optimized EC2 scheduling) 🎬 Max 100MB video, 100 subtitle lines 🐛 Active iteration - Bugs & UX improvements ongoing This is not a commercial service; it’s a personal, educational project exploring what’s possible with modern AI and cloud architecture. 😎 💭 Why I’m Sharing I’d love feedback from builders, creators, and educators: • What use cases am I missing? • What would make AI dubbing more valuable to you? • How can this better serve global content creation? I deeply respect the work by teams like ElevenLabs, Descript, and Papercup - BAK-Voice is my personal exploration, focused on accessibility and affordability for individual creators. 📺 Demo: https://lnkd.in/ggFp7nas 💬 Feedback: support@bakvoice.com Your feedback will directly shape v2. 🤠 Fun fact: The name BAK-Voice is inspired by the Hindi word "Bakwaas"(meaning "nonsense") - because every serious build needs a little humor. BAK doesn’t stand for anything, but there is a reason I chose those three letters… I’ll let you guess 😉
Kaushal Agarwal
Maker

The "One-Shot" Fallacy: Why I Rebuilt My AI Product 🛑

I built an AI video dubber. It worked like magic. And that was the problem.
A month ago, I launched BAK-Voice with a simple promise:
Upload video ➡️ Get translated video.

But as I started using it myself, I realized a painful truth about "One-Shot" AI:
The "Black Box" creates anxiety. 📉

You upload. You wait 30 minutes. You get the result. ❌
If one sentence has the wrong emotion... ❌ If one timing is off... You have to re-run the entire process. That’s not a tool; that’s a slot machine. 🎰 I realized that Automation without Control = Frustration. So, I tore down the workflow and built BAK-Voice Studio.

👇 The Pivot: From "Generation" to "Collaboration"
I stopped trying to replace the editor and started empowering them. The new vision? AI creates the rough draft. You create the final cut.

✨ New Features in Studio (V2):

✂️ Surgical Editing: Don't like how the AI said that one line? Just select that segment, edit the text, and hit regenerate. No need to re-process the whole video.

🐉 The "Goku" vs. "Vegeta" Protocol: I added model selection for different needs. 💎 (Yes, I’m a DBZ fan).

🎚️ Granular Audio Control:
• Keep Original Audio: Perfect for songs or untranslatable moments.
• Music Removal: Toggle background music removal per segment.
• Live Preview: Hear the audio before burning it into the video.

🏗️ The Architecture: Still running on a solo-built AWS serverless stack (Lambda, DynamoDB, SQS).

🎁 This is a personal educational project, not a SaaS startup. New users get 3 FREE video processing jobs. (Up to 5 mins, 100MB each). Enjoy unlimited studio editing! 😁

💡 The Lesson for Builders We are obsessed with "Zero-Shot" and "One-Click" AI. But users don't just want speed; they want agency. The future isn't AI doing it for us; it's AI doing it with us.

Try the new workflow: 🔗 https://bakvoice.com

I need your brutal feedback: Does the "Edit -> Regenerate" loop feel intuitive? What controls are still missing from your dream dubbing studio?
Let me know in the comments! 👇

Honest Note: To keep my wallet alive 💸, I won't be able to run this for more than a few weeks without getting free hashtag#AWS credits. 😉

(P.S. For those asking, "BAK-Voice" still comes from the Hindi "Bakwaas" meaning nonsense/rubbish. Because if you can't laugh at your own code, you're doing it wrong. 😉)

Kaushal Agarwal
Maker