Machine Talk

What happened?

This week I just wanted to try out the Speech to Text system I wrote about last week. A few years ago, OpenAI published an open source model called Whisper. The reason it interested me the most was because the model could be used offline in your own pc. Right now, the problem is that the end player would need either an NVIDIA GPU or a Mac with an M series chip. I was surprised to find that the model works pretty well. I was able to get it to work in a few minutes, and it transcribed my voice with a decent accuracy very fast.

Since my game would only use code-like messages, I only needed to create a simple parser that would take the text and convert it into a command, and a reader system that could check the type of utterance the player had given and build the command. I did a small flowchart where I defined the different utterances and combinations the game could understand. I found that I needed between three and for utterances to build a command with all the necessary info to proceed.

With that in mind I coded the phrase analysis system, considering the possibility that a code could be made of two words. After a couple of hours I had a system that could know if the player was saying a code or not. I still haven't connect the said code to any in-game logic, but I can already see that the Speech to Text system could be implemented for user testing. I think the interactions adds a lot to the game experience, and creating a written tex-based option is really simple.

After that I returned to coding the clients and guards behavior trees. Next week I want to have more complex scenarios happening in the store, and for that I need to keep developing the character’s controllers and AI.

What’s Wrong?

I still need to solve the core loop of this game. I have a clear idea of what it should be like, but from concept to implementation there is always a gap. I must give the players an experience that can be both fun and engaging as well as boring and frustrating.

Someone more experience would’ve told me to address this first. I’ve been doing many things for the sake of learning and seeing the game take shape, but I still haven’t developed the founding interaction that should make it work.

The incoming weeks should be about that. Making this game playable.