Release 2024-07-22
| Component | Version |
|---|---|
| Web portal (miarecweb) | 2024.7.22.8 |
| Recorder (miarec) | 2024.7.9.1 |
| Screen recorded (miarec_screen) | 2024.6.2.0 (not updated) |
| Live Monitoring (miarec_livemon) | 0.1.0.183 (not updated) |
| Screen capture client | 1.1.0.80 (not updated) |
New features
LLM-based Topic
In this release, we added Large-Languge-Model based Topic Analysis, which is a huge step up from the previous keyword-based technology.
First, it is more accurate as the AI is looking at a whole conversation rather than individual keywords. Second, it is easier to configure the topics. You just need to explain in plain English what certain topic means.
The following screenshot shows how to configure the topic.
In call details, the detected topics are shown with a list of key phrases that best represent the topic in the conversation:
AI Assistant populates automatically options in Custom Fields
If AI Task is used for categorizing calls into one of categories (for example, reason for the call), it is useful to allow AI Assistant to create new categories on flight.
With this software release, when AI Assistant creates a new category, such a category is automatically saved into custom field options.
This has multiple benefits:
-
When analyzing new recordings, the list of previously created categories is automatically injected into AI Prompt, so AI can select from the existing categories instead of creating many duplicates like "Cancel book reservation" and "Cancel a book reservation"
-
Users can select the existing options in a dropdown list when seaching recordings
Add support of extracting mutiple custom fields in one AI Prompt
Add "Raw RTP" passive recording interface
Add "Max call duration" setting to Raw RTP passive recording interface
This option is useful for recording long-standing continuous RTP streams, which can stay up for hours.
When maximum call duration is reached, the recording of the current file will be completed and, if RTP stream is still active, a new call record will be created.
With such a mechanism, a recording will be split on many small files rather than keeping everythign in a single very long file.
Add support of select listening interface explicitely for passive recording
Previously, the application was listening on all network interfaces by default.
Add support of extracting Twilio Flex task attributes into custom fields
Add ability to ignore transcript during hold periods
Currently, this is supported for Twilio Flex integration only.
This allows to exclude the Greeting / Music-on-hold portion of the call when analyzing call with Auto QA module.
Improve user experience when editing evaluation form
Now, you can create a form in one shot.
And you can re-order questions easily using drag'n'drop.
Extend replication to other data types
The following data can be replicated now:
- Topics,
- Ingegration,
- Tags,
- Speech Engines,
- SAML 2.0 Identity Providers,
- Clients,
- AI Tasks,
- Data Redaction Rules
Improve user experience in AI Prompt Playground
The results of a test are esier to review.
Improve access control to AI Prompt Playground
Tenant users can access AI Prompt Playground when they have "Edit" permission for Resource "AI Tasks".
Previously, it was required to grant "View" permission to "Speech Engines - Generative AI". This is not required anymore.
Add AI Prompt Library
Now, it is possible to save example AI Tasks in Prompt Library, which is visible to tenant users.
Tenant users can create own AI Tasks based on those example, using Clone option.
Add new events to Processing Queues
Now, you can subscribe queue to the following events:
- Transcription completed
- Redaction completed
- Repliction sent
- Replication received
This helps to reduce complexity of workflows. Previously, it was necesary to configure jobs to push explicitely to certain queues.
Now, the events are pushed to such a queue implicitely.
Add new role permissions for more granular control of access to Speech Analytics data
New permissions include:
- Call attributes - Custom fields
- Call attributes - Sentiment score
- Call attributes - Summary
- Call attributes - Topics
- Call attributes - Transcript
Add quick action "View only" in Role edit form
This allows administrator to configure the read-only role a lot faster than before.
Minor changes
Allow non-admin users to see Metaswitch-specific metadata in call details (User Name and Extension)
Improve user auto-provisioning for Twilio Flex integration
In this software release, the application can intelligently differentiate between Worker, Queue and external parties.
Improve user auto-provisioning for call scenarios with 3+ participants
Display "AI Insights" extracted data under "Trancript" tab rather than "AI Insights" tab
Display custom fields under "Call Info" tab rather than "Custom Fields" tab
Improvements to interaction handling
- Improve performance when retrieving calls that are part of the same interaction
- Reduce noise in logs when replicating calls that are part of the same interaction in parallel
Improve performance of S3 storage target
Improve performance of Storage Limits Enforcement job
Add "External Party" column in Recordings list
Extend call metadata in export, file relocation and replication jobs
All the important call attributes are included in call metadata file.
Fix inconsistent order in evaluation report list on Advanced Search page
Previously, it was sorged in Ascending order (oldest first). Now, it is sorted in Descending order (newest first), which is more consistent with the rest of the product features.
Improve handling of jobs with continuous schedule
Previously, the jobs were triggered only when queue was non empty. Now, they are additionally triggered when there are due work items, for example, re-try of the failed operation.
Decrease DB conection pool size in web application from 20 to 5
This change deals with potential issue when too many DB connections are opened under high load.
This is crititical for large deployment only.
[Ansible playbook] Optimize provisioning of web application under Apache web servers
Instead of having one process (mod_wsgi) with 15 internal threads, the application is not deployed with N processes and 5 threads per process, where N is equal to a number of vCPUs on the server.
[Ansible playbook] Enable DB and Redis connection pool recycle
The connections to DB and Redis servers are automatically recycled after 60 minutes.
This allows to deal with situation when TCP connections are stall (never closed) due to network issues.