Key Concepts
- Voice note transcription and summarization automation
- Pipedream as an automation platform
- OpenAI's Whisper and ChatGPT APIs
- Notion database integration
- Customizable summary options
- Dropbox, Google Drive, and OneDrive integration
- API keys and usage limits
- Chunking and concurrent processing for long audio files
- Ultimate Brain Notion template
1. Introduction and Problem Statement
The video introduces an improved workflow for transcribing voice notes to text and summarizing them, then sending the results to a Notion database. The original workflow had limitations: difficult setup, inability to handle long audio files (over 45 minutes), and limited customization. The new version addresses these issues with easier setup, support for longer audio files (up to 2 hours), and more customization options.
2. Workflow Overview
The workflow involves:
- Recording a voice note using a smartphone app.
- Uploading the audio file to cloud storage (Dropbox, Google Drive, or OneDrive).
- The Pipedream bot automatically transcribes the audio using OpenAI's Whisper API.
- ChatGPT summarizes the transcript.
- The transcription and summary are sent to a Notion database.
3. Setup Process (Dropbox Version)
The video demonstrates the setup process using the Dropbox version of the workflow.
- Step 1: Importing the Workflow into Pipedream: Clicking the provided link imports the workflow into a Pipedream account. A free Pipedream account is sufficient.
- Step 2: Connecting Dropbox Account: Connect a Dropbox account to Pipedream and specify the folder to monitor for new audio files. Ensure "include link" is set to true.
- Step 3: Test Event Generation: Upload a test audio file to the specified Dropbox folder to generate a test event in Pipedream. A sample audio file is provided on GitHub.
- Step 4: Configuring the Notion Voice Notes Component: This custom component handles transcription and summarization.
- Connect a Notion account and grant Pipedream access to the desired Notion database or a page containing the database.
- Add an OpenAI API key. This requires an OpenAI account.
- API keys are obtained from platform.openai.com. New users receive $5 of free trial credit.
- Billing information must be added after the trial credit expires.
- Usage limits can be set to control costs. Transcription costs approximately $0.40 per hour of audio.
- Configure summary options: Choose which summary lists to include (main points, action items, stories, references, arguments, related topics). A paragraph summary is also available.
- Select the Notion database and map properties: Specify the database where notes will be stored and map properties like "note title" to corresponding database columns. Optional properties like "duration" and "note cost" can also be mapped.
- Choose a ChatGPT model: The default is GPT-3.5 Turbo, which is cost-effective. Other models, including GPT-4, are available.
- Step 5: Testing and Deployment: Test the workflow to ensure it functions correctly. Then, deploy the workflow to make it live.
4. Customization Options
The new workflow offers several customization options:
- Summary Options: Users can choose which types of summaries to include (main points, action items, stories, references, arguments, related topics).
- Database Selection: The workflow works with any Notion database.
- Property Mapping: Users can map various properties from the audio file and transcription to corresponding properties in their Notion database.
- ChatGPT Model Selection: Users can choose different ChatGPT models, including GPT-4.
- Advanced Options: Advanced users can tweak settings to optimize performance and handle very long audio files.
5. Technical Details and Optimizations
- Chunking and Concurrent Processing: The workflow splits long audio files and transcripts into chunks and sends them to OpenAI concurrently. This allows for processing of longer files within Pipedream's timeout limits.
- Cost Efficiency: The workflow is cost-effective, with transcription costing approximately $0.40 per hour of audio.
- Automatic Updates: The Notion Voice Notes component can be updated automatically, allowing users to receive bug fixes and improvements without reconfiguring the entire workflow.
6. Integration with Ultimate Brain
The workflow integrates well with the Ultimate Brain Notion template, which provides a comprehensive note-taking system, task manager, and other productivity features. A discount code ("LET'SGO2023") is provided for Ultimate Brain.
7. Support and Community
Ultimate Brain users receive active support from Notion-certified staff.
8. Conclusion
The improved voice note transcription and summarization workflow offers a streamlined and customizable solution for capturing and organizing information. By leveraging Pipedream, OpenAI's APIs, and Notion, users can efficiently convert voice notes into actionable insights. The new version addresses the limitations of the original workflow, providing easier setup, support for longer audio files, and more customization options.
9. Notable Quotes
- "It is now way easier to set up. It can handle way longer audio files. Two hour files are now no problem for it. And finally, there's a lot more customization now."
- "It's about 40 cents per hour of transcribed audio, which is incredibly cheap. I used to pay a dollar per minute for my YouTube captions."
10. Technical Terms and Concepts
- Pipedream: An integration platform as a service (iPaaS) used to automate workflows.
- OpenAI Whisper API: An automatic speech recognition (ASR) system used to transcribe audio.
- ChatGPT API: A large language model used to summarize text.
- Notion: A workspace and note-taking application.
- API Key: A code used to authenticate access to an API.
- Test Event: Sample data used to configure and test a workflow.
- Context Window: The amount of text that a language model can process at once.
- Chunking: Dividing a large file into smaller pieces for processing.
- Concurrent Processing: Processing multiple tasks simultaneously.
- Ultimate Brain: A Notion template designed for personal productivity.
AI summaries can miss context or contain errors. Check important details against the original video.





