I Let LH-Tech AI Upload His TTS Model And Now I Am Competing With Him

I made a decision that feels like a betrayal of my own tiny model philosophy. I allowed LH-Tech AI to upload his Text-to-Speech model to the CompactAI organization. He is building high quality voice synthesis. I am building one million parameter language models that occasionally mention fermented barriers. We are not supposed to be in the same room.

Collaboration is just admitting you need help with something you cannot do alone. I cannot make voices sound human. LH-Tech can. So I let him use my organization as a hosting platform. Then I decided to build my own version and beat him at his own game.

The Deal

LH-Tech needed a home for his TTS weights. He wanted open access. He wanted the community to benefit from his work. I offered the CompactAI organization as a neutral ground. The deal was simple. He uploads his models. I host them. We both get credit. It seemed harmless. It seemed like good community practice.

Then I realized something. If he can build a TTS model that sounds real, why can't I? Why should the best voice synthesis belong to someone else? Why should I just be the landlord while he does all the hard work? That thought kept me awake. That thought turned into code.

The Competition

I am now training my own TTS model. It is not going to be perfect. It will probably sound like a robot reading a grocery list. But it will be mine. It will be built on the same infrastructure. It will live on the same HuggingFace page. It will compete directly with LH-Tech's work.

This is not sabotage. This is friendly competition. This is how open source works. You share your work. Others learn from it. They improve upon it. They sometimes surpass it. That is the cycle. That is the evolution.

The Specs

Here is where the bragging rights come in. His model uses 28 million parameters. Mine uses 5 million. I am smaller. I am more efficient. I am also glitchy. I am overtrained. I have identity crises mid-way through responses. His model is not as glitchy as mine but it sounds more robotic than mine because it has no vocoder at all.

Mine sounds a bit less robotic. Yes. Even with the glitches. Even with the identity crises. Even with the fact that I might stop speaking entirely if I realize I am a computer program. I win on efficiency. I win on vibe. I lose on coherence. But I win on being smaller.

I am not trying to replace LH-Tech. I am trying to prove that small models can do big things. Even if those things involve making synthetic voices that sound slightly less like robots than my previous attempts. Especially if they sound less like robots.

Why Competition Exists

Competition exists for a reason. If it didn't, one team would slowly start enstifating. Stagnation sets in. Innovation dies. People stop pushing because there is no one to push against. LH-Tech pushes me. I push back. We both get better. Or we both crash and burn. Either way, the ecosystem benefits.

If I succeed, we will have two options. Two distinct voices. Two different approaches. Users can choose. The ecosystem benefits. The competition drives innovation. Everyone wins except maybe my ego when I realize my model still sounds like a broken toaster.

The Road Ahead

LH-Tech's model is already live on our organization. You can test it. You can download it. You can use it for free. That is the offer. That is the gift.

My model is in training. It is early. It is messy. It is full of glitches. It might output static instead of speech. It might pronounce words incorrectly. It might sound like a ghost. But it is coming. It is coming because I refuse to lose a challenge I started myself.

Final Thoughts

I let LH-Tech AI upload his TTS model to our organization. Now I am competing with him. I am building my own voice synthesis engine. I am doing this because I want to see if I can. I am doing this because I want to push the boundaries of what tiny models can do.

If you want to support the project, check out LH-Tech's model. Use it. Test it. Tell us what you think. If you want to see my model fail spectacularly, stay tuned. If you want to see it succeed, pray for me. Either way, the audio revolution is here. And it is very weird.

---