Interrogator

Game allows player to use voice as a main input tool. To achieve great quality of speech to text, a base version of openai's whisper model is used. Model: https://openaipublic.azureedge.net/main/whisper/models/ed3a0b6b1c0edf879ad9b11b1af5a0e6ab5db9205f891f668f8b0e6c6326e34e/base.pt Repository with license: https://github.com/openai/whisper The content seen in the game is not live-generated. The player's input is embedded into a vector through a small model with 278M parameters. Model: https://huggingface.co/sentence-transformers/paraphrase-multilingual-mpnet-base-v2 Then, the game uses Faiss to match the player's input to known, predefined inputs. Each of those describes a reaction and change in the state of the game. Repository and License: https://github.com/facebookresearch/faiss Icon was made with local diffusion model.