How to transcribe audio to text with Whisper

Whisper transcribes audio to text with high accuracy, in many languages. Run on your server, it is private and with no per-minute costs.

1. Install

apt-get install -y python3-pip ffmpeg
pip install openai-whisper

2. Transcribe a file

whisper recording.mp3 --language English --model small

The result is saved as text and subtitles (.srt).

3. Choose the model

• tiny / base: fast, run on CPU, lower accuracy.
• small / medium: a good balance.
• large: the best accuracy, but needs a GPU for speed.

For large volumes

Transcribing many files or using the large model needs a GPU, see when you need a GPU and the GPU plans. With hourly billing, you start a GPU only for the transcription batch.

Automation

You can wire transcription into an automated flow with n8n: you upload a file, you get the text.