Tweet
Although the client opted out, sharing my experience in case anyone else has a similar client request.
My Prompt
Can you help me find a way to automate picking up new mp3 files from a folder and summarizing the call in text and creating a Lead with it in EspoCRM?
The Problem
Automatically convert call recordings (MP3/M4A files) and with it create a new Lead in EspoCRM, extracting details like names, loan amounts, home values, addresses, etc. and populating the corresponding fields
The Solution
Claude AI created a Python script that:
Download python 3.1.3, Free Groq account, ffmpeg (3 files from the zp file), create 2 folders, execute a few commands (Windows cmd) to install whisper, requests, watcher, and other functions
Challenges
Audio files failed processes until I added the 3 files from ffmpeg to the same folder the python script resided. My 1st recording was too low quality for Whisper to transcribe correctly. Total garbage. I used a different laptop to play the call and just used Windows built-in recorder (mpa) then used VideoLAN to convert to mp3. You can make the script to accept both but I wanted to mimic what the client would want to do/not do.
Lessons
You can see it filled in the 1) first name, 2) loan amount, 3) transcription, and 4) summary, The loan amount which was actually one of her last sentences, but you can see the summary -doesnt- have the 41,000.
This was my first pass last night, to be sure some tweaks would make it awesome. After the technical hurdles were solved I was happy to see how well it worked with EspoCRM!
Happy to answer question or send my notes if it helps you.
My Prompt
Can you help me find a way to automate picking up new mp3 files from a folder and summarizing the call in text and creating a Lead with it in EspoCRM?
The Problem
Automatically convert call recordings (MP3/M4A files) and with it create a new Lead in EspoCRM, extracting details like names, loan amounts, home values, addresses, etc. and populating the corresponding fields
The Solution
Claude AI created a Python script that:
- Monitors a folder (C:\AudioFolder) for new audio files
- Transcribes the audio using OpenAI's Whisper (runs locally on your PC - free!)
- Extracts lead information using Groq AI API (free tier)
- Names, email, phone
- Home value, loan amounts
- Property address
- Loan type, age/birth
- Generates a summary of the call
- Creates a lead in EspoCRM via the Lead Capture API
- Moves processed audio files to a processed/ subfolder with timestamps
- Continues to 'watch' for any new files dropped in that folder and immediately processes them
Download python 3.1.3, Free Groq account, ffmpeg (3 files from the zp file), create 2 folders, execute a few commands (Windows cmd) to install whisper, requests, watcher, and other functions
Challenges
Audio files failed processes until I added the 3 files from ffmpeg to the same folder the python script resided. My 1st recording was too low quality for Whisper to transcribe correctly. Total garbage. I used a different laptop to play the call and just used Windows built-in recorder (mpa) then used VideoLAN to convert to mp3. You can make the script to accept both but I wanted to mimic what the client would want to do/not do.
Lessons
You can see it filled in the 1) first name, 2) loan amount, 3) transcription, and 4) summary, The loan amount which was actually one of her last sentences, but you can see the summary -doesnt- have the 41,000.
This was my first pass last night, to be sure some tweaks would make it awesome. After the technical hurdles were solved I was happy to see how well it worked with EspoCRM!
Happy to answer question or send my notes if it helps you.