commands/sysadmin/linux-desktop/ai-setup/setup-speech-to-text.md

metadata

description: Check installed STT apps and suggest installations including local Whisper
tags:
  - ai
  - stt
  - whisper
  - speech-recognition
  - audio
  - project
  - gitignored

You are helping the user set up speech-to-text applications including local Whisper.

Process

1. Check currently installed STT apps

   • System packages: dpkg -l | grep -E "whisper|speech|voice"
   • Python packages: pip list | grep -E "whisper|speech|vosk"
   • Check ~/programs/ai-ml/ for installed apps

2. Suggest STT installation candidates

   Whisper (OpenAI) - Recommended:

   • Best quality, local inference
   • Multiple model sizes available
   • Multilingual support

   Other options:

   • Vosk - Lightweight, offline
   • Coqui STT - Mozilla's solution
   • SpeechNote - Simple GUI
   • Subtitle Edit - Video subtitling
   • Subtld - Automatic subtitles

3. Install Whisper (local)

   Method 1: Using pip (simple)

   pip install openai-whisper
   

   Method 2: Using conda (recommended)

   conda create -n whisper python=3.11 -y
   conda activate whisper
   pip install openai-whisper
   

   Install dependencies:

   # For audio processing
   sudo apt install ffmpeg
   pip install setuptools-rust
   

4. Install faster-whisper (optimized)

   pip install faster-whisper
   

   • Uses CTranslate2 for faster inference
   • Lower VRAM usage

5. Install WhisperX (advanced)

   pip install whisperx
   

   • Includes alignment and diarization
   • Better timestamps

6. Download Whisper models

   • Models are downloaded automatically on first use
   • Sizes: tiny, base, small, medium, large
   • Suggest based on VRAM:
     • < 4GB: tiny or base
     • 4-8GB: small or medium
     • 8GB+: large

7. Test installation

   whisper audio.mp3 --model base --language en
   

8. Install GUI options

   Whisper Desktop:

   • Check if available as AppImage or Flatpak

   Subtitle Edit:

   sudo apt install subtitleeditor
   

   Custom GUI:

   • Suggest installing gradio-based Whisper UIs

9. Create helper script

   • Offer to create ~/scripts/transcribe.sh:

     #!/bin/bash
     whisper "$1" --model medium --language en --output_format txt
     

10. Suggest workflows

• Real-time transcription
• Batch processing
• Video subtitling
• Meeting transcription

Output

Provide a summary showing:

• Currently installed STT applications
• Whisper installation status and model sizes
• GPU acceleration status
• Suggested models based on hardware
• Example commands for transcription
• GUI options available
• Helper scripts created