Skip to main content

OpenAI Whisper

Whisper is OpenAI's speech recognition model, known for high accuracy across multiple languages.

Setup​

1. Get API Key​

Whisper uses the same API key as OpenAI's other services:

  1. Go to OpenAI Platform
  2. Navigate to API keys
  3. Copy your API key

2. Configure Environment​

STT_PROVIDER=whisper
OPENAI_API_KEY=sk-...your-api-key

Configuration Options​

# Required
STT_PROVIDER=whisper
OPENAI_API_KEY=sk-...

# Optional
WHISPER_MODEL=whisper-1           # Default model
WHISPER_DEFAULT_LANGUAGE=en       # Language hint

Pricing​

MetricCost
Per minute$0.006
Per hour$0.36
3-min session (~1 min speech)~$0.006

Limitations​

  • No real-time streaming - Audio must be fully recorded before transcription
  • Max file duration - 2 hours per file
  • Batch processing - Slight delay while processing

Best Practices​

Optimize Audio Quality​

  • Use a good microphone
  • Minimize background noise
  • Ensure clear speech

Handle Silence​

  • Consider voice activity detection
  • Don't send silent audio segments

Troubleshooting​

"Invalid API key"​

  • Check your OpenAI API key
  • Ensure billing is set up

Slow transcription​

  • Whisper processes in batch, not real-time
  • Consider Deepgram for lower latency