SpeakLoop

AI-powered voice tutor for practicing foreign languages through real conversation (MVP).

About

SpeakLoop is a desktop application for practicing conversational foreign language with an AI partner. Hold Space to speak, release to get a response — the app transcribes your speech, sends it to an LLM, and reads the reply aloud.

Tech Stack

• GUI — Tkinter
• STT — faster-whisper (Whisper small by default)
• LLM — local GGUF model via llm_server/ or LM Studio
• TTS — Kokoro (hexgrad/Kokoro-82M)
• Python 3.11+

Installation

git clone https://github.com/yourusername/speakloop.git
cd speakloop

python -m venv .venv
.venv\Scripts\activate

pip install -r requirements.txt

For the LLM server — separate dependencies:

pip install -r llm_server/requirements.txt

For CUDA-enabled llama-cpp-python, see llm_server/README.md.

Configuration

All settings are in config.py:

# LLM backend selection
LLM_BACKEND = "local_server"   # recommended
# LLM_BACKEND = "lm-studio"   # if using LM Studio

# Path to the GGUF model file (for local_server)
EXTERNAL_MODEL_PATH = "models/llama-3.2-3b-instruct-q4_k_m.gguf"

# Language pair
NATIVE_LANGUAGE = "Russian"
TARGET_LANGUAGE = "English"

Running

python main.py

With LLM_BACKEND = "local_server" the server starts automatically. With LLM_BACKEND = "lm-studio" start LM Studio first.

Controls

• Space (hold) — record speech
• ESC — quit

Project Structure

speakloop/
├── main.py          — GUI, thread orchestration
├── stt.py           — Speech-to-Text (Whisper)
├── llm.py           — LLM client (OpenAI-compatible)
├── tts.py           — Text-to-Speech (Kokoro)
├── config.py        — all configuration
├── models/          — GGUF model files
└── llm_server/      — standalone process for local LLM
    ├── server.py
    ├── requirements.txt
    └── README.md