AI Voice-to-Text Telegram Assistant

High-precision transcription bot for voice messages, audio files, and video notes.

Client: Personal Tool Year: 2025 Duration: 1 week

Technologies Used

PythonTelegram Bot APISoniox AI

Challenge

Listening to long voice messages or video notes in meetings, noisy environments, or simply for archiving purposes is time-consuming. I needed a tool that could handle various Telegram media formats (OGG, MP3, MP4), bypass file size limitations, and provide accurate transcriptions for over 60 languages without storing sensitive data.

Solution

I developed a secure Python bot integrated with the Soniox AI engine. The system implements strict validation logic for file size (25MB) and duration (15 min). It features an asynchronous pipeline that downloads media, uploads it to the AI engine, retrieves the text, and intelligently chunks long responses to fit Telegram's 4096-character limit. Crucially, it includes a self-cleaning mechanism that deletes files from both the local server and the API provider immediately after processing.

Key Features

Multi-Format Support
Smart Text Chunking
Auto-Delete Security
User Access Control

Project Process

Phase 1: Setup

Configuring Telegram Bot API and setting up the Soniox authentication wrapper.

Phase 2: Core Logic

Implementing file handling (download/upload) and format validation (Voice/Audio/Video).

Phase 3: AI Integration

Connecting to Soniox Async API and handling polling for transcription status.

Phase 4: Optimization

Adding text splitting for long messages and error handling for timeouts.

Results

60+ Languages Supported

<10s Processing Speed

80% Transcription Accuracy