Telegram Audio Transcription Bot

Description

Introducing the Telegram Audio Transcription Bot: a cutting-edge solution that uses the telebot Python library in tandem with OpenAI's GPT-3.5-turbo and Whisper-1 models. It's designed to transcribe voice messages, refine these transcriptions using the GPT-3.5-turbo model, and then send back polished results to the user. The bot's multi-language support is exemplary; it automatically detects user language preferences and delivers translated system prompts and error messages for an enhanced user experience. Prioritizing user data security, the bot encrypts individual OpenAI API keys in the database.

Live Version: Experience the bot's capabilities first-hand at https://t.me/lmewhisperbot.

Changelog

• Version 2.1.0:

  • Integrated new DB with event logging.

• Version 2.0.0:

  • Integrated multi-language support with automatic language detection.
  • Introduced new commands: /changelanguage allowing users to switch between supported languages and /changekey to update OpenAI API keys.
  • Enhanced data security with encryption of OpenAI API keys within the database.
  • Revamped system prompts and error messages for intuitive user interactions.
  • Strengthened error handling and integrated detailed logging for streamlined troubleshooting.
  • Optimized overall bot performance.

Requirements

• Python 3.6+
• telebot library
• requests library
• openai library
• cryptography library
• mysql-connector library
• ffmpeg for audio conversion

Installation

pip install pyTelegramBotAPI requests openai cryptography mysql-connector-python

Note: ffmpeg must be installed separately. Please follow the official ffmpeg installation guide.

Setup

1. Secure an API key from OpenAI on their developer site.
2. Generate a Telegram Bot token using Telegram's BotFather guide.
3. Clone this repository.
4. Configure MySQL and construct the needed database and tables. The Database Setup section provides thorough details.
5. In the root directory, construct a config.json following the structure below:

{
    "encryption_key": "YOUR_GENERATED_ENCRYPTION_KEY",
    "openai_api_key": "your_openai_api_key",
    "bot_token": "your_telegram_bot_token",
    "db_config": {
        "host": "YOUR_DB_HOST",
        "user": "YOUR_DB_USER",
        "password": "YOUR_DB_PASSWORD",
        "database": "YOUR_DB_NAME"
    }
}

Tip: For generating an encryption_key, utilize the cryptography library in Python. Details are available in the Generating an encryption_key section.

6. Add or modify supported languages via the languages.json in the root directory.

Database Setup

1. Activate the MySQL server.
2. Implement the SQL commands below to set up the necessary database and table:

CREATE DATABASE TranscriptionBotDB;

USE TranscriptionBotDB;

	
CREATE TABLE `users` (
  `chat_id` bigint NOT NULL,
  `encrypted_openai_api_key` blob,
  `language` varchar(5) NOT NULL DEFAULT 'en',
  `nome_colonna` timestamp NULL DEFAULT CURRENT_TIMESTAMP,
  PRIMARY KEY (`chat_id`),
  UNIQUE KEY `chat_id` (`chat_id`)
);

CREATE TABLE `interactions` (
  `ID` int NOT NULL AUTO_INCREMENT,
  `ChatID` bigint NOT NULL,
  `Date` datetime NOT NULL DEFAULT CURRENT_TIMESTAMP,
  PRIMARY KEY (`ID`)
);

Note: The preferred_language defaults to 'en' (English). Users can adjust this setting through the /changelanguage command.

Usage

1. Start the bot with the command: python main.py.

2. On Telegram, find your bot using its username to begin a chat.

3. The bot recognizes the commands: /start, /help, /changekey, and /changelanguage:

   • /start & /help: Presents a warm welcome and essential instructions.
   • /changekey: Allows users to update their OpenAI API key.
   • /changelanguage: Enables users to select from a list of supported languages.

4. The bot is also proficient at managing voice messages, ensuring they are transcribed and enhanced using the OpenAI models.

Functions

• remove_phrases(text): Omit specific phrases from text.
• generate_corrected_transcript(temperature, system_prompt, audio_file): Refines transcribed content through the GPT-3.5-turbo model.
• send_welcome(message): Responds to the /start and /help commands.
• get_voice_message(message): Manages voice messages, transcribing and processing them.
• change_api_key_command(message): Enables users to modify their OpenAI API key stored in the database.

Generating an encryption_key

For ensuring the utmost security of OpenAI API keys in the database, make use of Fernet symmetric encryption:

from cryptography.fernet import Fernet
key = Fernet.generate_key()
print(key.decode())

Incorporate this generated key as the encryption_key in your config.json.

Disclaimer

The bot operates using OpenAI's GPT-3.5-turbo and Whisper-1 models. It's important to note that usage may lead to associated costs. Always check OpenAI's pricing for detailed information.

License

This project adheres to the terms of the MIT license.