Project Riko is a anime focused LLM project by Just Rayen. She listens, and remembers your conversations. It combines OpenAI’s GPT, GPT-SoVITS voice synthesis, and Faster-Whisper ASR into a fully configurable conversational pipeline.

tested with python 3.10 Windows >10 and Linux Ubuntu

• 💬 LLM-based dialogue using OpenAI API (configurable system prompts)
• 🧠 Conversation memory to keep context during interactions
• 🔊 Voice generation via GPT-SoVITS API
• 🎧 Speech recognition using Faster-Whisper
• 📁 Clean YAML-based config for personality configuration

All prompts and parameters are stored in config.yaml.

OPENAI_API_KEY: sk-YOURAPIKEY history_file: chat_history.json model: "gpt-4.1-mini" presets: default: system_prompt: |  You are a helpful assistant named Riko.  You speak like a snarky anime girl.  Always refer to the user as "senpai".  sovits_ping_config: text_lang: en prompt_lang : en ref_audio_path : D:\PyProjects\waifu_project\riko_project\character_files\main_sample.wav prompt_text : This is a sample voice for you to just get started with because it sounds kind of cute but just make sure this doesn't have long silences. 

You can define personalities by modiying the config file.

pip install uv uv pip install -r extra-req.txt uv pip install -r requirements.txt

If you want to use GPU support for Faster whisper Make sure you also have:

• CUDA & cuDNN installed correctly (for Faster-Whisper GPU support)
• ffmpeg installed (for audio processing)

python main_chat.py

The flow:

1. Riko listens to your voice via microphone (push to talk)
2. Transcribes it with Faster-Whisper
3. Passes it to GPT (with history)
4. Generates a response
5. Synthesizes Riko's voice using GPT-SoVITS
6. Plays the output back to you

• GUI or web interface
• Live microphone input support
• Emotion or tone control in speech synthesis
• VRM model frontend

• Voice synthesis powered by GPT-SoVITS
• ASR via Faster-Whisper
• Language model via OpenAI GPT

MIT — feel free to clone, modify, and build your own waifu voice companion.