Speech Generative AI Integration

Supported for CLOUD DEPLOYMENTS – On-Premise Deployments are supported if installed by Eleveo Support Engineers. Please contact your Eleveo representative to discuss a custom deployment if you require an on-premise installation.

What is it

Eleveo offers a Speech Generative AI package that is installed on a separate, dedicated, server. The Generative AI server creates a summary of the conversation (based on the transcription created by Eleveo Speech Recognition), and answers predefined questions about the conversation that were created by users(in the Automated Rules module). Quality Management then uses the answers to those user-created questions to score the conversation using Automated Rules. A similar approach is used to determine Topics and to provide an AI Rating for each conversation.

Combining the transcriptions provided by Eleveo Speech Recognition with Speech Generative AI and Automated Rules allow for 100% of conversations to be assessed and scored by Quality Management.

The solution is provided for on-premise and cloud deployments. Feature availability may vary based on your installation.

Multilingual Support

As of version 10.0 Eleveo Speech Generative AI supports multiple languages.
The system will generate summaries and other select outputs in the language defined by an administrator. Users can use the selected language when creating Automated Rules for Automated Quality Management.

Languages Tested and Verified For Generative AI Outputs

Eleveo tests the following langaages as part of the regular development process. Other languages are supported, but should be validated on a case-by-case basis before being deployed to a producion environment.

• Arabic

• Czech

• English

• Polish

• Spanish

Output By Language

• Summaries – Language of transcription as defined in configuration

• Insights – Highlights and Next Steps – Language of transcription as defined in configuration

• AI Rating Explanation – Language of transcription as defined in configuration

• Topics – English Only

• AI Grading – English Only (numerical value)

• Flags – English Only

• Automated Quality Management – Automated Rules – Language of transcription as defined in configuration

Known Limitations

• Please be aware that given the nature of large language models the output may not always be provided in the language of choice, the system may occasionally provide responses in English. This is especially the case for transcription files that contain two languages, for example when the conversation starts in one languages and switches later on to a second languages.

• Settings must be adjusted for Arabic language in order for the vLLM to respond reliably.

• When multiple languages are used across a conversation and there are multiple transcription files the language of the last transcript will be used by the system.

High-level Architecture Overview

Speech Generative AI is installed as an add-on to Quality Management and must be configured. This feature analyzes transcription files provided by the Speech Recognition Service and answers user-defined questions about the transcription. Quality Management combines the score with other variables/values based on the user-defined Automated rules. The combined scores are visible within the Conversation Explorer details pane as part of the automatic reviews results.

Chats and emails are not processed by Eleveo Speech Generative AI. Only transcriptions generated by Speech Recognition are sent for processing.

Detailed Visualization of Architecture

Speech Generative AI is dependent on Transcription files provided by the Speech Recognition server. Once the transcription file is available, the system sends it for processing.

The request is sent to the Speech Generative AI Server (marked by the label vLLM in the diagram below) along with any additional requests that will be processed by the Generative AI server (vLLM). Additional requests include; predefined Topics, specific questions associated with the AQM scoring, a summarization request, and any additional configuration required.

Responses from the Speech Generative AI server are stored in the database and made available to be displayed on the Conversation Explorer.

Supported Integration Use Cases

• Conversation Explorer – Display summaries created by the Generative AI within the Details Pane. Summaries are generated based on the transcription provided by Eleveo Speech Recognition.

• Quality Management – Provide responses to questions defined by users as part of Automated Rules.

Unsupported features:

• Email and Chats are not supported by Speech Generative AI

• Video calls are not processed at this time

• Historical media/recordings are not processed - only new recordings are processed by default

Supported / Unsupported Scenarios

• All new recordings that are processed by Speech Recognition (and have a transcription generated by Speech Recognition) are supported.

• Historical media/recordings are not processed.

• Archived media is ignored by Speech Recognition and Speech Generative AI and will not be (re)processed.

What Is Supported