CVG 1.45.0 (11-Feb-2025)

We are happy to announce major milestone with the general availability of VIER Speech-to-Text, some exciting enhancements in Smart Dialog and a new integrated conversational AI platform.

VIER STT now offers a high-performance, secure solution featuring a 50% reduction in transcription latency, elimination of duplicate word repetitions, enhanced transcription instructions, and increased capacity for large-scale projects—all hosted within our EU Data Center for top-notch security. In parallel, Smart Dialog has been upgraded with features like direct background audio control, an optional finalization prompt for post-dialog insights and streamlined handling of DTMF inputs. Last but not least, with Dify we have integrated a new conversational AI platform into CVG.

These dynamic updates embody a decisive leap forward, setting the stage for a future where voice communication is both rapid and secure, while seamlessly adapting to the ever-evolving digital landscape.

VIER Speech-to-Text Now Generally Available

We are excited to announce the General Availability (GA) release of VIER Speech-to-Text (STT). Following an extensive beta period, we have implemented a series of significant enhancements based on customer feedback and intensive monitoring. These improvements ensure a robust, scalable, and high-performance solution for our users.

Key Enhancements:

• Halved Transcription Latency: We’ve reduced the time required for transcriptions by 50%, delivering a smoother and more responsive experience for voicebot users.

• Eliminated Duplicate Word Repetitions: The issue where word sequences were sometimes mistakenly repeated at the end of a transcription has been completely resolved.

• Instruction prompt: When configuring a speech service profile you can give the VIER STT model instructions for transcribing the audio. This instructions can help achieve more accurate, context-aware transcriptions that are aligned with the intended content and style.

• Increased Capacity for High-Volume Projects: Our capacity upgrade now enables us to effectively support large-scale, high-volume customer projects — an essential prerequisite for our GA release.

Security & Deployment: Local Data Center Operation

VIER STT is hosted locally within our EU Data Center (not yet available in US), ensuring compliance with the highest security standards and safeguarding sensitive information.

We are confident that these enhancements will deliver a superior user experience and meet the demanding needs of our diverse customer base. Thank you for your continued support as we move VIER Speech-to-Text into its next phase of innovation.

Smart Dialog

Smart Dialog continues to evolve as the go-to solution for using simple LLM-based bots without the need for a separate Conversational AI platform. In this release, we have introduced several Smart Dialog enhancements that deliver a more immersive and streamlined user experience.

Background Audio Option

Previously available via API since release 1.41, background audio playback can now be directly controlled through Smart Dialog bot templates. You can specify an audio file to be played in a continuous loop, enriching the conversational ambiance and delivering a more immersive interaction.

Finalization Prompt for smoother integration with downstream data processes

A new optional “Finalization Prompt” is executed automatically after each dialog concludes. You can use this prompt e.g. to generate a summary of the conversation, which can then be captured as custom data via function calling. This feature provides enhanced post-dialog insight and enables smoother integration with downstream data processes.

Use "custom data" and "key" in your prompt

The results of the “Finalization Prompt” are generated by AI and saved via an internal function call. Use the wording “custom data” and “key”, e.g. as in the example above, to ensure that this function call is made and the result is saved.

Ignore DTMF Input

To streamline the dialog experience, Smart Dialog bots will now ignore DTMF inputs. Previously, each pressed DTMF digit triggered a new LLM request, often leading to confusing interactions. This update ensures that unintentional key presses no longer disrupt the conversation flow.

New Conversational AI Integration: Dify (Beta)

Dify’s mission is to make the creation and deployment of generative AI applications easily accessible. By providing intuitive tools, collaborative environments and cutting-edge models, Dify enables developers and creatives to innovate quickly and efficiently. With the new Dify integration in CVG, we are extending these capabilities to the field of voice-activated phone services.

CVG’s new bot template seamlessly converts a text-based Dify bot into a voicebot optimized to interact with phone services. This integration leverages the robust CVG platform to enable clear and reliable voice communication.

A simple genAI-based voicebot built with Dify and CVG could look just like this:

Dify integration is beta

The Dify integration is currently available in CVG as a beta version. We value your feedback via support@vier.ai as it drives future improvements and helps us create the ultimate Voicebot development experience using the conversational AI tool of your choice.

Smaller Improvements

Playing and downloading Recordings

In the Dialogs tab of projects, users can now play and, for the first time, download recordings. Please note that playlists are no longer supported.

Allow searching by provisioning number

Projects can now also be found by phone numbers in their provisioning blocks, rather than just platform phone numbers.

Improve rendering of provisioning blocks

Provisioning blocks now always display their phone number range in the project settings, the description has moved to a tooltip. Also, the differing parts of the range are now highlighted.

Improve dialog API behavior

The behavior of the Dialog API has been imporved to better meet your expectations:

• Retrieving a dialog now returns an empty response instead of 404 if the dialog exists, but there are no entries (e.g. after provisioning or with includeType filters).

• It is no longer possible to attach custom data to a dialog after a dialog has been deleted.

Fix /call/play blocking the request

When the queue is empty, requests to /call/play would block until the playback is started.

Fix terminated events without session

For calls that disconnect immediately, the bot’s terminated endpoint was still invoked even though there was never a session request beforehand.