How To Use Whisper (OpenAI)

Getting Started (5 minutes)

Whisper is OpenAI's speech-to-text AI, but it's not a standalone app you download. It's built into other tools. Here's how to access it:

Option 1: Use it through ChatGPT (easiest)

Open the ChatGPT mobile app or go to chatgpt.com in your browser.
Tap the headphone icon in the bottom right (mobile) or the microphone button (web).
Speak. ChatGPT transcribes what you say using Whisper, then responds.

Option 2: Use it through an API or third-party app

Apps like Otter.ai, Descript, and MacWhisper use Whisper under the hood. Install one of those if you need long-form transcription (meetings, interviews, podcasts).
For developers: Get an API key from OpenAI, then upload audio files via the Whisper API endpoint. Check OpenAI's docs for pricing and limits.

Transcribe meetings or interviews. Record audio on your phone, drop the file into a Whisper-powered app like MacWhisper, and get accurate text in seconds. Works with 99 languages.
Generate subtitles for videos. Upload your video's audio track and Whisper spits out timestamped captions. Way faster than typing them manually.
Convert voice memos to text. Dictate notes while driving or walking, then turn them into readable text later. Whisper handles accents, background noise, and informal speech better than most transcription tools.

Clean audio = better results. Whisper is good with background noise, but if you're transcribing something important (like a legal deposition), record in a quiet room or use a decent mic.
Use timestamps if available. Some Whisper-powered apps (like Descript) show you exactly where each word appears in the audio. Click a word in the transcript and the audio jumps to that moment - huge for editing interviews.
Check for technical terms. Whisper is trained on a massive dataset, but it sometimes stumbles on niche jargon, brand names, or acronyms. Proofread before publishing.

Real-time live captions that must be perfect. Whisper is fast (near real-time), but for critical accessibility use cases like live broadcasts, dedicated captioning services with human review are safer.
Transcribing heavily overlapping speech. If multiple people talk over each other constantly, Whisper struggles to separate voices. It works best with clear, turn-based conversation.