Here is the revised README.md for your LLMTTS project, strictly professional and without emojis:

---

LLMTTS (Large Language Model Text-to-Speech)

LLMTTS is a Python-based real-time system that streams responses from a Large Language Model (LLM) and converts them into speech using Microsoft Edge TTS and pygame. It supports Thai and English, streaming text generation, dynamic audio buffering, and voice fallback.

---

Features

• Real-time text generation and audio playback
• Seamless voice fallback mechanism across multiple TTS voices
• Thai language support with optional tokenization
• Text-to-speech conversion using Microsoft Edge neural voices
• Integration with OpenTyphoon LLM via OpenAI-compatible API

---

How It Works

1. Sends a user prompt to the OpenTyphoon LLM.
2. Streams generated text token by token.
3. Buffers text until a sentence is complete or a threshold is reached.
4. Converts buffered text to audio using edge-tts.
5. Plays audio in real-time using pygame while streaming continues.

---

Requirements

Install the dependencies with:

pip install openai edge-tts pygame thai-tokenizer

Additional system requirement: ffmpeg must be installed and available in your system path for edge-tts to function.

---

Environment Setup

Set the environment variable for your API key:

os.environ["OPENAI_API_KEY"] = "<your-api-key>"

Make sure to point to the correct base URL if using a custom or third-party LLM service:

base_url="https://api.opentyphoon.ai/v1"

---

Voice Support

The system uses the following Microsoft neural voices (in priority order):

• th-TH-NiwatNeural
• th-TH-PremwadeeNeural
• en-US-ChristopherNeural

It will attempt fallback in this order if a voice fails during audio generation.

---

Directory Structure

LLMTTS/
├── llmtts.py             # Main application script
├── requirements.txt      # Optional, list of dependencies
└── README.md             # Documentation

---

Usage Example

messages = [
    {
        "role": "user",
        "content": "วันนี้กินข้าวกับอะไรดี"
    }
]

The model responds in real-time, and the audio is generated and played sentence-by-sentence.

---

Notes

• The Tokenizer from thai_tokenizer is initialized for future use in sentence segmentation or domain-specific parsing.
• The queue-based design ensures audio playback does not block streaming or processing.
• This script is suitable for voice assistants, chatbots, or language learning tools.

---