1Stream Enterprise Required
1stream Enterprise Required
This feature is available for only 1stream Enterprise licensed users and higher plans. If you don't have this and would like this feature to be available, please reach out to your Sales Representative to upgrade user licensing.
OpenAI Consumption Required
This process will cause consumption-based charges to start in your OpenAI Platform account. Consult your bill after a day or two to determine if your settings need to be adjusted, or if this is not a right fit for you.
Requirements
You must have the following in order to proceed:
- An OpenAI Platform account
- Call recording enabled for whichever calls you want to analyze.
Create the API Key for OpenAI
- Login to OpenAI Platform
- In the top left click your current project and then Create new project
- Give it a name like 1Stream and click Create
- Go to Dashboard > API Keys
- Click Create new secret key
- Enter in 1Stream for the name again and click Create
Copy down the key received, you'll use that in the next section.
Setup the 1Stream Integration
Now that we have our Azure account setup for this, we can setup the 1Stream side.
- Login to your 1Stream admin portal and go to Administration > Manage Organization.
- Scroll down to the Connected Accounts and CRM Settings section and click the green + sign in the top right.
- In the Add New CRM Platform pop up window, select Google Speech and then click Save.
- Enter in your API Key, then click Save.
Modifying the Prompt and Model
We give you a basic prompt template which provides consistent transcription results, however to get the best results you'll want to modify the prompt to best fit your company's needs. You have full access to modify the prompt and the model that will be used when sending call recordings to OpenAI.
To modify the AI Prompt, you'll want to use an external markdown editor such as Visual Studio Code or StackEdit. You can follow general AI prompt engineering guides such as Text generation and prompting - OpenAI API to figure out what exactly to write.
Some companies may find the full transcription to be too much and that they only want the summary, sentiment, and next steps, so you can completely remove that section if you so wish.
For the AI Model, we specifically utilize the Chat Completions API to generate the transcription and summary. When looking at the Models list by OpenAI, you specifically need a model that supports Audio input and the Chat Completions endpoint.
The following are the current recommended models:
- GPT-4o mini Audio
- AI Model: gpt-4o-mini-audio-preview
- This is the default set when you configure the integration
- You can set this to gpt-4o-mini-audio-preview-2024-12-17 to prevent it from automatically changing to the next release.
- We estimate $0.50-$0.75 per hour of talk time
- GPT-4o Audio
- AI Model: gpt-4o-audio-preview
- This provides higher accuracy at a much higher cost
- You can set this to gpt-4o-audio-preview-2025-06-03 to prevent it from automatically changing to the next release.
- We estimate $1.50-$2.50 per hour of talk time
Our cost estimates are just that, estimates, so we implore you to monitor your billing for the first few days after implementation to ensure it's within your expectations.
Next Steps
Follow our guide below on setting up 1Stream to send specific call recordings to OpenAI.