Demo Audio Transcription

In this tutorial, you’ll learn how to transcribe a short audio clip using OpenAI’s Whisper API. We’ve prepared a trimmed MP3 of the first five minutes of Steve Jobs’ Stanford commencement speech for this demo.

Prerequisites

• Python 3.7+
• An active OpenAI API key
• openai Python package (pip install openai)
• IPython for in-notebook audio playback (pip install ipython)

1. Play Audio Locally

Before sending the file to Whisper, verify playback in an IPython environment:

import IPython

file_name = "data/jobs.mp3"
IPython.display.Audio(file_name)

2. Transcribe with Whisper

Whisper currently offers the whisper-1 model for speech-to-text. Set your API key in the environment, then transcribe:

import openai
import os

# Load your OpenAI API key from environment
openai.api_key = os.getenv("OPENAI_API_KEY")

file_name = "data/jobs.mp3"
with open(file_name, "rb") as audio_file:
    transcript = openai.Audio.transcribe("whisper-1", audio_file)

print(transcript.text)

export OPENAI_API_KEY="your_api_key_here"

3. Next Steps: NLP Pipelines

Once you have the raw transcript, you can feed it into large language models like GPT-3.5 Turbo or GPT-4 to:

• Summarize the speech
• Generate Q&A bots
• Classify or analyze sentiment
• Extract key topics

4. Run Whisper Locally

If you prefer not to use the API, you can run Whisper on your machine via the open-source repository:

References