Whisper is OpenAI’s Automatic Speech Recognition (ASR) system. The system has been trained on about 700,000 hours of supervised data, both multilingual and multitask, collected from the Internet. Thanks to this training, accomplished with a diverse and massive set of data, Whisper manages accents, background noise, and technical language with impressive ease. It also performs transcription in numerous languages and translates these languages into American English. 

Implemented as an encoder-decoder transformer, Whisper’s architecture is an uncomplicated, end-to-end approach: it breaks input audio into 30-second pieces, which are converted into a log-Mel spectrum and sent through an encoder; the decoder is trained to anticipate the proper text caption, combined with special tokens that direct the single model to undertake language identification, multilingual speech transcription, phrase-level timestamps, and speech translation to-English.

In November of last year we announced a beta version of the Vodia PBX that connects the telephone system to the beta version of the OpenAI realtime API. If your organization prioritizes data sovereignty and on-premises processing, Vodia also supports the deployment of Whisper AI within your dedicated infrastructure. This enables you to maintain full control over your transcription processes, ensuring sensitive call data remains securely within your network boundaries.

Configuration Steps​

To set up a self-hosted Whisper AI instance, follow these steps:

1. Access Tenant Settings
   
   • Log in to your Vodia tenant account.
   • Navigate to General Settings.
2. Provide Deployment Details
   
   • Enter the URL of your Whisper AI deployment.
   • Provide the authentication credentials
     
     • Username
     • Password

3. Save and Connect
   
   • Save the configuration to establish a secure connection between your Vodia PBX and your local Whisper AI instance.

Accessing Call Transcriptions​

To view the transcribed content, simply log in to your user portal, navigate to the 'History' section, select the desired call, then examine the 'call content' area.

Hardware Requirements​

To ensure optimal performance when running Whisper AI on your own hardware, refer to the official hardware requirements outlined in the OpenAI Whisper GitHub repository.

Benefits of On-Premises Deployment​

• Data Sovereignty: Keep sensitive call data within your network.
• Full Control: Manage and customize transcription processes according to your organization's needs.
• Enhanced Security: Ensure compliance with internal and regulatory security standards.

Thanks to the integration of our PBX with Whisper API, it’s easy to transcribe calls - you can use the OpenAI Whisper or host your own Whisper server for true data privacy. To get started with on-premise transcription, see our Whisper AI on-premise setup documentation or review a self-hosted Whisper integration example. If you're interested in a cloud-based setup, follow our cloud-based call transcription guide.

Now that we’re supporting real-time AI API integration with OpenAI, we’re also looking at integrating with more AI providers, so we can provide seamless AI integration within workflows. We’d love to tell you all about it - reach out to us at sales@vodia.com or call +1 (617) 861-3490 (United States), +61 2 7201 0788 (APAC), or +49 30 555 78749 (Europe).