OpenAI Whisper Beginner

Whisper is OpenAI's open-source speech recognition model. Trained on 680,000 hours of multilingual data, it delivers near-human accuracy across 99 languages. You can run it locally for free or use the OpenAI API for production workloads.

Installing Whisper Locally

Whisper can be installed via pip. It requires Python 3.8+ and ffmpeg for audio processing.

Bash

# Install Whisper and its dependencies
pip install openai-whisper

# Install ffmpeg (macOS)
brew install ffmpeg

# Install ffmpeg (Ubuntu/Debian)
sudo apt install ffmpeg

# For faster inference, install faster-whisper
pip install faster-whisper

Whisper Model Sizes

Whisper comes in several model sizes. Larger models are more accurate but slower and require more VRAM:

Model	Parameters	VRAM	Relative Speed	Best For
tiny	39M	~1 GB	~10x	Quick testing, low-resource devices
base	74M	~1 GB	~7x	Simple transcription tasks
small	244M	~2 GB	~4x	Good balance of speed and accuracy
medium	769M	~5 GB	~2x	High accuracy for most use cases
large-v3	1550M	~10 GB	1x	Maximum accuracy, multilingual

Basic Transcription with Python

Python

import whisper

# Load the model (downloads on first use)
model = whisper.load_model("base")

# Transcribe an audio file
result = model.transcribe("meeting.mp3")

# Print the full transcript
print(result["text"])

# Access word-level timestamps
for segment in result["segments"]:
    print(f"[{segment['start']:.1f}s - {segment['end']:.1f}s] {segment['text']}")

Using the Whisper CLI

Whisper also includes a command-line interface for quick transcription:

Bash

# Basic transcription
whisper audio.mp3 --model base

# Specify language and output format
whisper audio.mp3 --model medium --language en --output_format srt

# Translate non-English audio to English
whisper french_audio.mp3 --model large --task translate

Using the OpenAI Whisper API

For production use, the OpenAI API provides a hosted Whisper endpoint with fast response times and no GPU requirement on your end:

Python

from openai import OpenAI

client = OpenAI()

# Transcribe audio via API
with open("meeting.mp3", "rb") as audio_file:
    transcript = client.audio.transcriptions.create(
        model="whisper-1",
        file=audio_file,
        response_format="verbose_json",
        timestamp_granularities=["word"]
    )

print(transcript.text)

# Access word-level timestamps
for word in transcript.words:
    print(f"[{word.start:.2f}s] {word.word}")

Faster Whisper

Faster Whisper uses CTranslate2 for up to 4x faster inference with the same accuracy:

Python

from faster_whisper import WhisperModel

# Load model with int8 quantization for speed
model = WhisperModel("large-v3", compute_type="int8")

segments, info = model.transcribe("audio.mp3", beam_size=5)

print(f"Detected language: {info.language} ({info.language_probability:.0%})")

for segment in segments:
    print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")

Performance Tip: For production workloads, Faster Whisper with int8 quantization on the large-v3 model gives you the best accuracy-to-speed ratio. It runs well on a GPU with 6+ GB VRAM or even on CPU for shorter audio files.

Try It Yourself

Install Whisper, download a sample audio file, and transcribe it using both the local model and the CLI. Experiment with different model sizes to see the accuracy-speed tradeoff.

Next: Cloud APIs →

← Introduction Cloud APIs →

Ready to Go Deeper?

Live instructor-led courses from our partners. Affiliate disclosure.

🧠

AI & ML Courses - 30% Off

Live instructor-led AI, machine learning, data science, and cloud courses for working professionals. Use code Limited30 at checkout.

Edureka

📊

DataCamp - AI & Data Science

Hands-on Python, machine learning, and AI courses with interactive exercises and real projects.

DataCamp

🎓

edX - Top AI Courses

University-level AI courses from MIT, Harvard, Stanford. Earn certificates that employers recognize.

edX