Hey everyone! I’m trying to set up a cool project with my Telegram bot. The goal is to take voice notes and turn them into text automatically. I’ve heard about this thing called Whisper from OpenAI that can do speech recognition, and I want to run it locally.
The problem is, I’m not sure how to make it work with N8N, which is what I’m using for my bot. Has anyone done something like this before? I’m kind of stuck on how to connect all the pieces.
I’d really appreciate any tips or advice on how to get started. Maybe there’s a specific node I should be using in N8N? Or do I need to write some custom code? Thanks for any help you can offer!
hey john, i’ve done smthing similar. u could try using the ‘Execute Command’ node in n8n to run whisper locally. just make sure u have whisper installed on ur machine first. then u can pass the audio file path to the command and grab the output. it’s not perfect but it works. good luck!
I’ve actually implemented something similar for a client’s project recently. Here’s what worked for me:
First, you’ll need to set up a local instance of Whisper. It’s pretty straightforward if you follow the GitHub instructions. Once that’s running, you can create an HTTP endpoint that accepts audio files.
In N8N, use the Telegram Trigger node to listen for voice messages. When one comes in, use the HTTP Request node to send the audio file to your local Whisper instance. The response will be the transcribed text.
From there, you can use N8N’s core functionality to process the text however you want - maybe send it back to the user, store it in a database, or trigger other actions based on the content.
One tip: make sure your audio preprocessing is solid. Whisper works best with clean audio, so consider using ffmpeg to normalize the audio before sending it to Whisper.
Hope this helps point you in the right direction. Let me know if you need any clarification on the steps.
I’ve experimented with a similar setup using N8N and Whisper. Here’s a potential approach:
Set up a webhook in your Telegram bot to receive voice messages. In N8N, use the Webhook node to catch these incoming messages. You’ll need to extract the file_id of the voice note.
Next, use the Telegram node in N8N to download the actual audio file using the file_id. Once you have the file, you can send it to a locally hosted Whisper instance via an HTTP Request node.
For Whisper, I’d recommend using the faster ‘base’ model unless you need high accuracy. The transcription result can then be sent back to the user or processed further in your workflow.
One challenge you might face is handling long voice messages, as they can timeout. Consider implementing a queueing system for larger files to process them asynchronously.
Remember to handle potential errors gracefully, as speech recognition isn’t always 100% accurate.