n8n Node: Audio to Text (Whisper)

This is an n8n node to transcribe audio files into text using local, on-device models from the Hugging Face Transformers.js library. It leverages the powerful Whisper model family by OpenAI.

The primary advantage of this node is that all processing happens directly within your n8n instance, ensuring data privacy and eliminating reliance on external cloud services for transcription.

Features

• Local Transcription: No data leaves your server.
• Multiple Model Options: Choose from various Whisper model sizes (tiny, base, small, medium) to balance speed and accuracy.
• Multilingual Support: Transcribe audio in numerous languages or let the model auto-detect the language.
• Translation Task: Directly translate audio from any supported language into English.
• Flexible Input: Works with audio file URLs or binary data from previous n8n nodes.

Prerequisites

This node requires n8n version 1.0 or later.

Installation

1. Navigate to your n8n's custom nodes directory.
2. Run the following command:

   npm install n8n-nodes-rckflr-audiototext

3. Restart your n8n instance.

Usage

After installation, you can find the "Audio to Text (Whisper)" node in the "Transform" category.

1. Audio Input: Provide a direct URL to an audio file or use an expression to reference binary data from a preceding node (e.g., {{$binary.data}}).
2. Model: Select the desired Whisper model. Smaller models are faster but less accurate. Multilingual models support various languages, while English-only models are optimized for English.
3. Language: (Optional) Specify the two-letter ISO code for the language of the audio (e.g., es for Spanish). If left empty, the model will attempt to auto-detect it.
4. Task: Choose between transcribe (to get text in the original language) or translate (to get the text translated to English).

The node will output a text field containing the transcribed text and a chunks field with a more detailed breakdown of the transcription segments.

License

MIT