Kindred Therapist AI

Kindred is an AI-powered emotional support tool that combines speech-to-text, large-language-model dialogue, and speech generation into a simple Flask web UI.

---

Requirements

• Hardware:
  • CUDA-compatible GPU (≥4 GB VRAM recommended)
• OS & Language:
  • Python 3.10+
  • Linux, macOS, or Windows 10+
• Libraries & Tools:
  • torch & torchaudio
  • transformers
  • flask
  • pygame
  • soundfile
• Other:
  • ffmpeg on PATH (for audio operations)
  • Access to Hugging Face models:
    • sesame/csm-1b (CSM speech generator)
    • meta-llama/Llama-2-7b-chat-hf (chat LLM)
    • facebook/wav2vec2-base-960h (speech-to-text)

---

Setup

1. Clone & enter repo

   git clone https://github.com/MohamedIbrahim404/Kindred.git
   cd kindred-therapist-ai

2. Create & activate virtual env

   python3.10 -m venv .venv
   source .venv/bin/activate   # Windows: .\.venv\Scripts\activate

3. Install Python deps

   pip install -r requirements.txt

4. Login to Hugging Face & pull models

   huggingface-cli login

---

Notes

GUI will probably be changed in the near future or entirley replaced with an online platform. I'm also working on optmizing the responses as for some reason the voice changes every response and the audio is never longer than 8 seconds.