Dictation & Dictation APIs - Version History

0.21.2

February 2026 - See Product Deployment Dates for more information.

• Fix for Unapplied Vocabulary Changes -  In previous releases, newly added vocabulary terms at the group (organization) level were not being recognized during dictation. The system was not applying the updated vocabulary correctly due to a profile error. This issue has now been fixed. (D-13974, 12/9/2025, 0.20.0-rc.7)
• Fix for Delayed Recognition of Newly Added Vocabulary - A delay was observed in which newly uploaded vocabulary was not immediately recognized during dictation. Although the vocabulary was correctly added to the users’ default topic (General Medicine), dictations performed under a different topic (Pathology) did not reflect the update in a timely manner. As a result, the new vocabulary was not applied during Pathology dictations, even though it was available for General Medicine. This has been fixed. (D-13884, 1/21/2026, nVoq Admin 21.0.0 b41 & profile server v1.3.7)
• Speech Recognition Accuracy Improvements - We’ve integrated recent enhancements to the speech recognition pipeline to further improve overall accuracy and consistency. As part of this update, rescoring behavior was reviewed and refined as needed to ensure recognition accuracy meets or exceeds previous performance. (S-18104, 2/5/2026,  ReDS v0.21.1)
• Fix for HTTP Dictations Stuck in Requeue Loop - We’ve improved the reliability of HTTP dictation start and reduced repeat processing in cases where account lookups are slow or unresponsive. Previously, some number of users could see a 2-3 second delay at the start of dictation, followed by “Problem getting account properties” and repeated re-queues that could continue until the job eventually timed out. With this update, the system allows a bit more time for the initial account check to complete and, if it still can’t, it returns a clearer failure reason so the Gateway can stop retrying and end the job promptly. (D-13993, 2/5/2026, ReDS v0.21.2, Gateway v2.0.2)
• Fix for MODIFYDICTATION Triggering 1008 Policy Violations - We fixed an issue that could cause WebSocket dictations to be rejected with error 1008 (policy violation) when running multiple dictations using MODIFYDICTATION. In some timing scenarios, a prior session could remain marked active after a MODIFYDICTATION transition, which could inflate the system’s per-user request count and trigger a 1008 rejection. With this update, the session is now properly cleared during close so request limits are evaluated correctly and subsequent dictations can proceed normally. A broader improvement to the underlying transition logic will be delivered separately under D-14085. (D-14046, 2/5/2026, ReDS v0.21.2)

  
  

20.3.0

December 2025

• Fix for "Lowercase" Substitution Not Working - The built-in substitution "lowercase" was not functioning as expected when used with direct text. Instead of modifying the text, the word "lowercase" appeared in the transcript. This issue has been resolved. Now, saying "lowercase" at the start of a dictation correctly removes the word and formats the next word with a lowercase initial letter. (D-13828, 8/1/2025, 20.3.0 b6)
• Fix for WebSocket Connection Delays - Users of the nVoq WebSocket API previously experienced long delays when establishing a connection, which sometimes resulted in audio not being processed. This issue has been resolved, ensuring that audio is consistently processed and output is always returned. (S-18496, 8/18/2025, 20.3.0 b15)
• Fix for Conversation Transcript Sometimes Not Returning - We resolved a problem in which the diarized transcript from the Conversations feature in Voice would sometimes not return, and spinner kept running. When this happened, the dictation showed a fatal error in Review & Correct. This update ensures Conversation results appear as expected and dictation errors no longer occur. (D-13846, 9/11/2025, 20.3.0 b33)
• Fix for Long Silence Causing HTTP Error 500 - We resolved a defect that occasionally caused a 500 Internal Server Error when the system reached the configured silence threshold. (D-13634, 9/15/2025, 20.3.0 b33)

20.2.0 Release

July 2025

Dictation

• Inaccurate Initial Transcriptions for Pathology and Radiology Dictations - Some dictations on the Pathology and Radiology topics were returning inaccurate transcriptions during initial processing. However, resubmitting the same audio through the same accounts often produced improved results. Testing revealed the issue was related to inconsistent boundary requests. We’ve updated the boundary request logic to help improve the accuracy of transcripts sent for review. With these updates, however, there is a chance that part of the audio from the start of one segment could be included at the end of the previous segment. (D-13833, 6/24/2025, 20.2.0 b16)
• Fix for Missing Words in Dictation Transcripts - In the previous release, some users experienced missing words in their dictation transcripts. This issue has been resolved by improving how the system determines when to segment audio during transcription, resulting in more accurate and complete transcripts. (D-13863, 7/9/2025, 20.2.0 b25)
• Increased Limit for Organization and Account Substitutions - In previous releases, users with more than 3,000 literal organization or account substitutions may have experienced some substitutions not working, as the system could only retrieve up to 3,000 of each from the server. In the 20.2.0 release that limit has been raised to 10,000 for both organization and account substitutions. This means the system can now support up to 20,000 substitutions in total, ensuring broader coverage for users with large substitution lists. (D-13874, 7/17/2025, 20.2.0 b32)

Dictation APIs

• Fix for WebSocket Observer API Not Supporting SSO with Wireless Microphones - The WebSocket Observer API was not properly handling Single Sign-On (SSO) requests for WebSocket dictation with nVoq Wireless Microphones due to an incorrect organization identifier. The API has been updated to use the correct identifier, and SSO with WebSocket dictation now functions as expected. No API documentation changes. (D-13448, 6/24/2025, 20.2.0 b16)