Speech-to-Text Transcription

Description

Comprehensive speech-to-text capabilities using multiple STT engines including Whisper (local/free), Google Speech, Azure Speech, and AssemblyAI. Record audio, transcribe files, real-time processing, speaker identification, timestamps, and multi-language support.

Features

• ✅ Multiple STT engines (Whisper, Google, Azure, AssemblyAI)
• ✅ Audio recording with voice activity detection
• ✅ File transcription (WAV, MP3, M4A, FLAC, OGG)
• ✅ Real-time streaming transcription
• ✅ Speaker diarization (identify who said what)
• ✅ Multi-language support (99+ languages)
• ✅ Timestamp and segment tracking
• ✅ Multiple output formats (text, JSON, SRT, VTT)
• ✅ Batch processing
• ✅ Audio enhancement (denoise, normalize)
• ✅ Video subtitle generation

Installation

Prerequisites

• Python 3.8+
• Microphone (for recording)
• API keys (for cloud engines)

Setup

1. Install Whisper (free, local):

   pip install openai-whisper
   

2. For cloud engines (optional):

   pip install google-cloud-speech azure-cognitiveservices-speech assemblyai
   

3. Audio dependencies:

   pip install sounddevice soundfile numpy
   

Usage

Record and Transcribe

# Record audio
python scripts/record_audio.py --output recording.wav

# Transcribe with Whisper (free)
python scripts/transcribe_whisper.py --file recording.wav

# Real-time transcription
python scripts/stream_whisper.py

Engine Comparison

Engine	Cost	Speed	Quality	Best For
Whisper	Free	Medium	High	Privacy, offline
Google	$1.44/hr	Fast	High	Real-time
Azure	$1/hr	Fast	High	Enterprise
AssemblyAI	$0.90/hr	Medium	Very High	Analysis

Examples

Example 1: Meeting Transcription

Record meeting, transcribe with speaker labels, export as markdown

Example 2: Voice Notes

Quick voice note recording and transcription to markdown

Example 3: Video Subtitles

Extract audio from video, generate SRT subtitles

Configuration

Choose Whisper model size:

• tiny/base: Fast, good for notes
• small/medium: Balanced
• large: Best accuracy, slower

License

MIT License - see LICENSE for details.

Author

Aaron Storey (@astoreyai)