Read the full submission
Project Description:
This project automates the transcription of long voice recordings from mentoring sessions into structured notes in Airtable. These recordings, often lasting up to three hours, are converted into text using Deepgram and processed for storage and analysis in Airtable. Given Airtable’s cell character limit of 100,000, the transcript is split into three manageable chunks and stored in separate fields.
The automation also generates summaries, categories, and tags for the transcripts using OpenAI, making the data more accessible and easier to navigate. This solution not only saves time but also ensures that valuable insights from mentoring sessions are stored systematically for future reference.
Achievements:
Efficient Handling of Long Transcripts:
- Overcame Airtable’s character limit by splitting transcripts into three chunks, each under 100,000 characters.
Accurate Transcriptions:
- Used Deepgram for high-quality transcription, ensuring the integrity of the original recordings.
Enhanced Data Organization:
- Stored structured data, including summaries, categories, and tags, in Airtable for easy retrieval and analysis.
Seamless Workflow:
- Automated the entire process from file upload to transcription, summarization, and database entry.
Tools Used:
Make :
- Orchestrates the workflow, connecting all tools and automating the steps.
Google Drive:
- Serves as the source folder for audio files uploaded from the user’s iPhone.
Deepgram:
- Provides high-quality transcription of the audio files.
OpenAI GPT Model:
- Generates transcript summaries, categories, and tags for better organization.
Airtable:
- Stores the transcript data, including chunked text, summaries, categories, and tags, in a structured format.
Methodology:
Monitor Folder for New Audio Files:
- The workflow begins by monitoring a designated folder in Google Drive for new audio files.
Download and Transcribe Audio:
- Audio files are downloaded and sent to Deepgram, which transcribes the recordings into text.
Split Transcript into Chunks:
- The transcript is divided into three chunks, each under 100,000 characters, using the Set Variable module in Make.
- Chunk 1: First 99,999 characters.
- Chunk 2: Second 99,999 characters.
- Chunk 3: Remaining characters.
Generate Additional Data:
- Transcript Summary: Created using OpenAI, providing a concise overview of the session.
- Category and Tags: Automatically generated by OpenAI to classify and organize the content.
Store Data in Airtable:
- Data fields include:
- Title
- Transcript_Summary
- Transcript_Chunk1
- Transcript_Chunk2
- Transcript_Chunk3
- Category
- Tags
- Transcript_Document (as TXT file)
- Date_Added
Save Transcript File:
- The full transcript is saved as a TXT file and uploaded into Airtable for backup and future use.
Move Processed Files:
- Once the process is complete, the original audio file is moved to another folder to avoid duplication.
Learnings:
Scalable Transcription Solutions:
- Learned how to handle long transcripts efficiently, ensuring compatibility with Airtable’s character limits.
Data Enrichment:
- Gained insights into enhancing raw transcription data with summaries, categories, and tags for better usability.
Workflow Optimization:
- Refined the workflow for handling large files, ensuring seamless automation without errors.
Improved Storage Practices:
- Implemented a system for organizing and backing up data, making it easily accessible for future reference.
Project Description:
This project automates the transcription of long voice recordings from mentoring sessions into structured notes in Airtable. These recordings, often lasting up to three hours, are converted into text using Deepgram and processed for storage and analysis in Airtable. Given Airtable’s cell character limit of 100,000, the transcript is split into three manageable chunks and stored in separate fields.
The automation also generates summaries, categories, and tags for the transcripts using OpenAI, making the data more accessible and easier to navigate. This solution not only saves time but also ensures that valuable insights from mentoring sessions are stored systematically for future reference.
Achievements:
Efficient Handling of Long Transcripts:
- Overcame Airtable’s character limit by splitting transcripts into three chunks, each under 100,000 characters.
Accurate Transcriptions:
- Used Deepgram for high-quality transcription, ensuring the integrity of the original recordings.
Enhanced Data Organization:
- Stored structured data, including summaries, categories, and tags, in Airtable for easy retrieval and analysis.
Seamless Workflow:
- Automated the entire process from file upload to transcription, summarization, and database entry.
Tools Used:
Make :
- Orchestrates the workflow, connecting all tools and automating the steps.
Google Drive:
- Serves as the source folder for audio files uploaded from the user’s iPhone.
Deepgram:
- Provides high-quality transcription of the audio files.
OpenAI GPT Model:
- Generates transcript summaries, categories, and tags for better organization.
Airtable:
- Stores the transcript data, including chunked text, summaries, categories, and tags, in a structured format.
Methodology:
Monitor Folder for New Audio Files:
- The workflow begins by monitoring a designated folder in Google Drive for new audio files.
Download and Transcribe Audio:
- Audio files are downloaded and sent to Deepgram, which transcribes the recordings into text.
Split Transcript into Chunks:
- The transcript is divided into three chunks, each under 100,000 characters, using the Set Variable module in Make.
- Chunk 1: First 99,999 characters.
- Chunk 2: Second 99,999 characters.
- Chunk 3: Remaining characters.
Generate Additional Data:
- Transcript Summary: Created using OpenAI, providing a concise overview of the session.
- Category and Tags: Automatically generated by OpenAI to classify and organize the content.
Store Data in Airtable:
- Data fields include:
- Title
- Transcript_Summary
- Transcript_Chunk1
- Transcript_Chunk2
- Transcript_Chunk3
- Category
- Tags
- Transcript_Document (as TXT file)
- Date_Added
Save Transcript File:
- The full transcript is saved as a TXT file and uploaded into Airtable for backup and future use.
Move Processed Files:
- Once the process is complete, the original audio file is moved to another folder to avoid duplication.
Learnings:
Scalable Transcription Solutions:
- Learned how to handle long transcripts efficiently, ensuring compatibility with Airtable’s character limits.
Data Enrichment:
- Gained insights into enhancing raw transcription data with summaries, categories, and tags for better usability.
Workflow Optimization:
- Refined the workflow for handling large files, ensuring seamless automation without errors.
Improved Storage Practices:
- Implemented a system for organizing and backing up data, making it easily accessible for future reference.
Relevant URLs:
- Deepgram Official Website
- Airtable Official Website
- Make (Integromat)
Conclusion:
The Voice Memo Transcription Automation project exemplifies how AI and automation can simplify the management of large volumes of audio content. By combining high-quality transcription with systematic storage and categorization, it transforms lengthy mentoring sessions into actionable and organized data.
The Mindvalley AI Mastery Program was instrumental in the success of this project. Through the program, I developed the skills to design workflows that merge advanced AI tools with practical use cases. This project demonstrates how automation can unlock the full potential of recorded insights, making them easily accessible and impactful for personal and professional growth.
