v2.61 | 05 August 2024
We are continuing to improve turn-taking with the Digital Person. In this release, we have improved the Digital Person’s ability to detect that it was interrupted and give the participant extra time to complete their turn. When the Digital Person is speaking and is interrupted by the user, the Digital Person will wait a half-second longer before speaking next time. A half-second is added each time the Digital Person is interrupted up to a maximum of three seconds. Each time the Digital Person is not interrupted while speaking a half second is removed from the wait time. You can enable this feature by selecting “Enable users to verbally interrupt the Digital Person” in your project configuration.
When Interruption and Streamed responses are enabled the Digital Person knows where it was interrupted and can continue from that point if asked.
Streamed responses are now out of beta! New projects will use this by default. You can opt-out by toggling this off in the Generative Conversation Base settings.
We fixed an issue where “Boost Expressiveness” was causing Digital Person speech to be cut short in both live deployments and in the video creator.
v2.60 | 22 July 2024
We are continuing to improve turn-taking with the Digital Person. In this release, we have improved the Digital Person’s ability to recover from verbal interruptions by adjusting the conversation history so the LLM understands it was interrupted.
We fixed an issue where typing into the chat box wouldn’t interrupt the Digital Person.
We fixed an issue where adding
@hidecards()would create an empty speech bubble in the chat transcript.We fixed an issue where empty speech bubbles were created in the chat transcript when the Digital Person was interrupted.
v2.59 | 8 July 2024
We’re updating how we handle billing and eliminating the need to sign in to the billing portal a second time. Users should now be able to manage their billing via the “Account & Team” page. We are rolling this feature out gradually.
We updated the content card skill to enable you to show the cards as the Digital Person speaks, rather than all at the start.
We added a Spanish Language coach to the AI Assistant Gallery on
Soul Machines Gallery | AI Assistants We made it more clear when Alex the Language Coach was in ‘tutor’ mode or ‘roleplay’ mode, by having the camera change.
We fixed an issue when the Digital Person would speak punctuation marks.
v2.58 | 25 June 2024
We’re updating how we handle billing and eliminating the need to sign in to the billing portal a second time. Users should now be able to manage their billing via the “Account & Team” page. We are rolling this feature out gradually.
We’re continuing to improve the experience related to turn-taking. This release, we’ve improved the naturalness of how the Digital Person’s speech is finished when interrupted by the user.
We fixed an issue for generative conversation streaming [beta] which caused the Digital Person to prefer replies of one sentence in longer conversations
Release v2.57 | 11 June 2024
We’ve improved the Digital Person’s turn-taking ability, resulting in fewer instances of the Digital Person speaking over the end-user. You can toggle this feature on and off per project by selecting “allow the user to verbally interrupt the Digital Person”. Please note this feature is in beta. Read more here.
We’re updating how we handle billing and eliminating the need to sign in to the billing portal a second time. We are rolling this feature out gradually.
We’ve increased the amount of conversation history that a Digital Person can recall
We enabled a 14 day trial for the Pro plan
We updated the syntax for the Pronunciation skill to use a
|instead of a-. Read more here
We fixed an issue where the transcript didn’t reflect the spoken content when using Generative Conversation
We fixed an issue in Dutch language which caused the Digital Person to exit after speaking the word "eerste"
Release v2.56 | 28 May 2024
We now support Deepgram for speech recognition. Deepgram offers excellent speech recognition abilities and boasts low latency.
We released a “Pronunciation” skill, which allows you to teach your Digital Person how to pronounce certain phrases phonetically.
Coming Soon: We’re improving the Digital Person’s turn-taking ability, resulting in fewer instances of the Digital Person speaking over the end-user. Once released, you can toggle this feature on and off per project.
The Voiceflow Skill now allows you to set your “Base URL”.
We’re updating how we handle billing and eliminating the need to sign in to the billing portal a second time. We are rolling this feature out gradually.
We’ve added new Google and Microsoft voices, and removed any that were no longer supported
We fixed an incompatibility between the Watson base conversation Skill and the Generative Conversation Fallback Skill.
We fixed a bug where updating your Organization name didn’t reflect in the interface.
Release v2.55 | 29 April 2024
We’ve made several updates to the Generative Conversation Base & Fallback.
Under “Generative Conversation Base” there a new toggle to match the conversation style to the behavior style chosen during set-up, which will be toggled off by default. When turned on, the chosen behavior style tags will be added to the experience.
We’ve renamed the “personality” field in the set-up to “Additional conversation style prompting.” Add more personality queues here to enhance your experience.
We’ve fixed a bug so when using the Generative Conversation Base & Fallback the Digital Person doesn’t add any interruptions by the user to the chat transcript. Additionally, all spoken greetings by the Digital Person are now included in the transcript.
Release v2.54 | 15 April 2024
The Generative Conversation Skill and Fallback have been updated so that only correctly formatted URLs with https:// can be used in the "Knowledge URLs" section. Without the correct formatting, the knowledge snippets won't be updated.
Release Version v2.53 | 1 April 2024
We’ve added support for Romanian, Polish and Swedish so your Digital Person can speak and understand even more languages.
We’ve added more tool tips to the Generative Conversation Skill & Fallback to help users better understand what fields inform different aspects of the experience. The Generative Conversation Skill can also now support larger amounts of data (more than 4096 bytes) when “streaming” is toggled on.
Skills
We've been working hard to make sure the skills in our platform align with customer needs and the shift towards Generative AI based conversations. With that in mind, the following skills are in the process of being deprecated. If you use any of these skills and have questions, please email [email protected].
Weather (Wolfram Alpha)
Twitter
Site Search
Google Form Survey
Spelling
Shopify
YouTube Channel
Greeter Base
Elegant Failure
Knowledge Base Fallback
Repeat
Facebook
Knowledge Base (Base corpus)
Joke
Contact
Translation (Microsoft Cognitive)
Math
Auto-Join Microsoft Teams
We’ve reduced the delay time when unmuting/muting a Digital Person with projects using Google’s STT.
A bug was fixed in projects using cinematic camera cuts so that URLs are pronounced correctly when using the @pronounce function.
We’ve fixed a bug so that prices with punctuation, specifically commas, are maintained as a single entity, where previously they were displayed and read as separate entities with incorrect spacing. Now, you can have your Digital Person discuss pricing with no issues!
Release Version 2.52 | 18 March 2024
We’ve added an improved response time by toggling on or off streaming in the Generative Conversation Skill & Fallback. This allows long answers to be broken up into multiple sentences that are delivered at one time to keep the conversation engaging. By default, this button is toggled on. You can toggle it off at any point under the Generative Conversation Base settings, beneath the “Max response length” field.
Our Skills UI just got a new look!
Streamlined Experience: Check out the separated “Conversation” and “Additional Skills” sections under “Knowledge” when creating or editing a project - they’re much easier to navigate!
Conversation Simplified: All projects must have a “Conversation” selected, but this can be updated easily based on your project’s needs.
Easy Selection: When choosing a conversation provider, you'll find them neatly categorized as "Featured" and "External Platform" for effortless selection.
Using Orchestration? Simply select the "Orchestration Skill" in the Conversation section.
Focused Skill Selection: When adding “Additional Skills”, only non-conversation skills will appear to make it even easier to enhance your project.
We’ve fixed a bug with the Digital Chief Experience Officer (DCXO) template so the Digital Person now responds with an error-handling prompt for the user to repeat themselves.
A bug was identified and fixed in the orchestration server when no NLP was selected.
Release Version 2.51 | 4 March 2024
We’ve added a BETA stock Digital Person named Kate who will be wearing glasses. Kate is not editable in the Comprehensive Blender Tool just yet, but we’re working towards the ability to accessorize Digital People, and this is a great first step! Check her out in your studio account.
We’ve removed the Knowledge Base Builder from the main menu as this product is being moved to end of life status. If you have an existing project in Knowledge Base Builder, you can still access the URL directly.
Legacy Digital People using Human OS 2.2 or lower will be removed from studio as they’re no longer supported.
Content Cards can now be used with the Solution Expert Template as well as Generative Conversation so you can add multi-modal content to your experience.
A bug with creating video assets has been fixed so video controls will remain visible and the entire script will be recorded.
Release Version 2.50 | 19 February 2024
We’ve updated the Content Card Skill so that it now supports markdown tables which can be used with the Generative Conversation Base.
Release Version 2.49 | 22 January 2024
We’ve added support for Bahasa Malay (Malaysian) for both STT & TTS on azure speech services.
We’ve added Automatic Iconic Gestures. You can enable this for your project to take advantage of automatic gesture insertion for any conversation backend. You toggle this feature on and off under “Other Behavioral Settings” on the “Create a Project Page.” Automatic Iconic Gestures enable your Digital Person to perform additional, more expressive gestures when speaking. If these gestures don’t work in your experience, leave this toggle off. Learn more about this in Behavior Adjustments and Settings.
Generative Conversation Base & Fallback skill
We’re constantly updating our Generative Conversation Documentation to include the latest features and best practices for set-up.
Resolved an issue to allow URLs to show as specified in the knowledge snippets when using the Generative Conversation Base & Fallback skill
We’ve fixed an error when using the Generative Conversation Base & Fallback skill where URLs with large amounts of text were not being scraped. All URLs will now scrape up to the first 40,000 tokens. You can turn this on or off by toggling “Convert URL data into optimized conversation content” under the “Generative Conversation Base” settings.
We’ve resolved a bug with Microsoft STT timing out so connection remains intact.
We’ve fixed a bug so Digital People viewed in preview mode are now connecting correctly to OpenAI.
Was this article helpful?
That’s Great!
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
Feedback sent
We appreciate your effort and will try to fix the article