Call analyzer with AssemblyAI transcription and OpenAI assistant integration

Video Guide

I prepared a detailed guide that showed the whole process of building a call analyzer.

Who is this for?

This workflow is ideal for sales teams, customer support managers, and online education services that conduct follow-up calls with clients. It’s designed for those who want to leverage AI to gain deeper insights into client needs and upsell opportunities from recorded calls.

What problem does this workflow solve?

Many follow-up sales calls lack structured analysis, making it challenging to identify client needs, gauge interest levels, or uncover upsell opportunities. This workflow enables automated call transcription and AI-driven analysis to generate actionable insights, helping teams improve sales performance, refine client communication, and streamline upselling strategies.

What this workflow does

This workflow transcribes and analyzes sales calls using AssemblyAI, OpenAI, and Supabase to store structured data. The workflow processes recorded calls as follows:

1. Transcribe Call with AssemblyAI: Converts audio into text with speaker labels for clarity.
2. Analyze Transcription with OpenAI: Using a predefined JSON schema, OpenAI analyzes the transcription to extract metrics like client intent, interest score, upsell opportunities, and more.
3. Store and Access Results in Supabase: Stores both transcription and analysis data in a Supabase database for further use and display in interfaces.

Setup

Preparation

1. Create Accounts: Set up accounts for N8N, Supabase, AssemblyAI, and OpenAI.
2. Get Call Link: Upload audio files to public Supabase storage or Dropbox to generate a direct link for transcription.
3. Prepare Artifacts for OpenAI:
   • Define Metrics: Identify business metrics you want to track from call analysis, such as client needs, interest score, and upsell potential.
   • Generate JSON Schema: Use GPT to design a JSON schema for structuring OpenAI’s responses, enabling efficient storage, analysis, and display.
   • Create Analysis Prompt: Write a detailed prompt for GPT to analyze calls based on your metrics and JSON schema.

Scenario 1: Transcribe Call with AssemblyAI

1. Set Up Request:
   • Header Authentication: Set Authorization with AssemblyAI API key.
   • URL: POST to https://api.assemblyai.com/v2/transcript/.
   • Parameters:
     • audio_url: Direct URL of the audio file.
     • webhook_url: URL for an N8N webhook to receive the transcription result.
     • Additional Settings:
       • speaker_labels (true/false): Enables speaker diarization.
       • speakers_expected: Specify expected number of speakers.
       • language_code: Set language (default: en_us).

Scenario 2: Process Transcription with OpenAI

1. Webhook Configuration: Set up a POST webhook to receive AssemblyAI’s transcription data.

2. Get Transcription:

   • Header Authentication: Set Authorization with AssemblyAI API key.
   • URL: GET https://api.assemblyai.com/v2/transcript/<transcript_id>.

3. Send to OpenAI:

   • URL: POST to https://api.openai.com/v1/chat/completions.
   • Header Authentication: Set Authorization with OpenAI API key.
   • Body Parameters:
     • Model: Use gpt-4o-2024-08-06 for JSON Schema support, or gpt-4o-mini for a less costly option.
     • Messages:
       • system: Contains the main analysis prompt.
       • user: Combined speakers’ utterances to analyze in text format.
     • Response Format:
       • type: json_schema.
       • json_schema: JSON schema for structured responses.

4. Save Results in Supabase:

   • Operation: Create a new record.
   • Table Name: demo_calls.
   • Fields:
     • Input: Transcription text, audio URL, and transcription ID.
     • Output: Parsed JSON response from OpenAI’s analysis.