Pre-Installed Whisper

Overview

Whisper is a family of state-of-the-art, open-source AI-powered speech recognition models developed by OpenAI. Whisper delivers leading performance in multilingual speech recognition, speech translation, language identification, and supports word-level timestamps, speaker diarization, and streaming transcription—with robust performance across diverse accents, background noise, and technical language.

Select a pre-installed Whisper Marketplace image and get a fully configured server that's ready to use immediately—no manual installation or complex setup required.

Supported Operating Systems

The Whisper Marketplace image is available for Ubuntu Server 24 LTS 64-bit.

Supported Products / Plans

The Whisper Marketplace image is currently available across the following plans:

GPU VPS
Professional GPU VPS - RTX Pro 2000
Professional GPU VPS - RTX A4000

GPU Dedicated Server
Basic Dedicated GPU Server - T1000
Basic Dedicated GPU Server - RTX 4060
Basic Dedicated GPU Server - RTX 5060

View compatible plans →

Whisper Version

Database Mart provides the v1.0.7 version of Whisper, pre-installed and ready to use.

What's Installed

When you select the Whisper Marketplace image, Database Mart automatically installs and configures:

Core Operating Environment

• Operating System: Ubuntu Server 24.04 LTS, pre-optimized for speech recognition workloads
• GPU Drivers & Acceleration: NVIDIA drivers and CUDA pre-installed for GPU acceleration (when applicable)
• Python Environment: Python 3.8+ with a dedicated virtual environment for dependency management

Whisper Core Components

• Whisper Model: Pre-trained model weights downloaded automatically. Available sizes include tiny (39M), base (74M), small (244M), medium (769M), large (1.55B), and turbo (798M)
• Audio Processing Tools: FFmpeg pre-installed for handling various audio formats (MP3, WAV, FLAC, etc.)
• Inference Engine: Pre-configured with OpenAI Whisper for local offline transcription

Inference Optimization Tools

• CPU/GPU Inference: Pre-configured to run on CPU for general use or GPU for faster transcription
• Multi-Language Support: Supports 98+ languages with automatic language detection
• Output Formats: Supports JSON, TXT, SRT, and VTT output formats

Installation Process

During server provisioning, Database Mart automatically performs the following steps:

1. Deploys the selected Linux operating system (Ubuntu Server 24.04 LTS).

2. Installs system dependencies, including FFmpeg for audio processing, Python 3.8+, and the pip package manager.

3. Creates a Python virtual environment and installs core dependencies:

   • openai-whisper (the official Whisper package)
   • PyTorch with CUDA support (for GPU acceleration) [citation:9]
   • Audio processing libraries

4. Downloads and configures the selected Whisper model weights from OpenAI's repository (models are stored in ~/.cache/whisper/).

5. Configures the default model and output settings for immediate use.

Everything is fully automated—your Whisper server is ready to use as soon as deployment is complete.

Getting Started After Deployment

After your server has been provisioned, follow these steps to begin using Whisper.

Step 1: Access the Server

Once deployment is complete, connect to your server via SSH using the credentials provided by Database Mart:

ssh administrator@your-server-ip

Step 2: Activate the Environment and Verify Installation

# Activate the pre-configured virtual environment
cd /opt/whisper
source venv/bin/activate

# Verify the installation
whisper --help

Expected output: Help menu showing available commands and options.

Step 3: Test Basic Transcription

To test the installation with a simple audio file:

# Make sure you have an audio file (e.g., test.mp3)
whisper test.mp3 --model base --language English

Step 4: Using Different Models

# Use the small model for better accuracy
whisper audio.mp3 --model small

# Use the large model for maximum accuracy (requires more VRAM)
whisper audio.mp3 --model large

# Use the turbo model for faster inference
whisper audio.mp3 --model large-v3-turbo

Step 5: Specify Output Format and Language

# Transcribe to English and output as SRT subtitles
whisper audio.mp3 --model base --task transcribe --language English --output_format srt

# Translate audio to English
whisper audio.mp3 --model base --task translate

# Output as JSON with timestamps
whisper audio.mp3 --model base --output_format json

Step 6: Python Script Usage

import whisper

# Load the model
model = whisper.load_model("base")

# Transcribe audio
result = model.transcribe("audio.mp3", language="English")

# Print the transcription
print(result["text"])

# Access segments with timestamps
for segment in result["segments"]:
    print(f"{segment['start']:.2f}s - {segment['end']:.2f}s: {segment['text']}")

Notes

Audio Format Support
Whisper supports various audio formats including MP3, WAV, FLAC, and M4A. For best results, use 16kHz mono audio.

Multilingual Support
Whisper supports 98+ languages and can automatically detect the language of the audio. You can also specify the language manually for better accuracy.

Performance Tips

• Use faster-whisper (optimized CTranslate2 version) for 4-8x speedup with minimal quality loss
• For GPU acceleration, ensure CUDA drivers are installed
• For large models, use --fp16 flag to reduce memory usage

Frequently Asked Questions

Q: What audio formats are supported?

Whisper supports common audio formats including MP3, WAV, FLAC, and M4A. FFmpeg is pre-installed to handle format conversion automatically.

Q: Does Whisper support multiple languages?

Yes. Whisper supports 98+ languages with automatic language detection. You can also specify the language manually with the --language parameter for better accuracy.

Q: Can I run Whisper offline without internet?

Yes, once the model is downloaded, Whisper runs entirely offline. The models are stored locally in ~/.cache/whisper/.

Q: What output formats are supported?

Whisper supports multiple output formats: TXT (plain text), SRT (subtitle format), VTT (WebVTT), and JSON (with timestamps and segments).