Release notes
Updated
New features, improvements and resolved issues for Real-time STT.
Releases
v7.2.3
Released on July 30, 2026
New features
Included in this release:
-
New webhook event for an expiring RTC Token
This version adds the
104 token will expireevent. The Notifications service sends this event to your server when the RTC Token used by the Real-Time STT engine has 30 seconds remaining before it expires. Agora does not send this event if the RTC Token's validity period is 30 seconds or less. This event provides advance notification only; it does not end the task or change the task's final status.
Upgrade notes
-
Updated idle timeout and maximum task lifetime rules
This version updates the rules for the
maxIdleTimeparameter: the valid range changes to[0,259200]seconds, and settingmaxIdleTimeto0disables automatic termination due to idle time.Independent of
maxIdleTime, every task also has a maximum lifetime, capped at 72 hours. When a task reaches this limit, Agora terminates it, even ifmaxIdleTimeis0. When a task exits because it reached the maximum lifetime, the102 agent leftevent returnsTask exceeded maximum lifetimein themessagefield. -
Transcription and translation results can return stable and interim segments together
Transcription and translation results can now contain both a stabilized prefix segment and a segment that may still change within the same update. To support this, the
transcriptandtranslationmessages add a newresults[]field that returns one or more segments. For a single segment,results[]has a length of1; for a mixed result,results[]can have a length greater than1.If your client uses the JSON protocol, iterate over
results[]to process each segment, and handle messages whereresults[]is an empty array. The Protobuf protocol fields are unchanged, but a single message can now also carry multiple segments. For field descriptions and examples, see JSON protocol data structure.
v7.2.2
Released on May 18, 2026
New features
Included in this release:
-
Support for additional languages
This version expands the supported language lists for real-time transcription and real-time translation. See Supported languages for the full list.
Upgrade notes
-
Updated target language configuration rules for real-time translation
This version updates the rules for configuring source and target languages in real-time translation, for both single-language and mixed-language input scenarios:
-
Single-language input: If you set the source language to a single language, the target language must be different, otherwise an error is returned. For example, if you set the source language to English, you cannot set the target language to English.
-
Mixed-language input: If you set the source language to mixed-language input, you can set the target language to one of the source languages. For example, if you set the source languages to Spanish and English, setting the target language to English translates both into English.
-
v7.2.1
Released on April 20, 2026
New Features
Included in this release:
-
Support for caption storage
This version adds support for writing recorded subtitles to standard S3-compatible object storage. You can configure parameters such as
endpoint,type, andproviderin thecaptionConfig.extensionParamsof the transcription request to connect to S3-compatible storage services such as MinIO, as well as some self-hosted object storage. See Record captions for details. -
Agent list API
This version adds a List Real-time STT agents API. Use it to retrieve real-time transcription and translation tasks that meet specified criteria by channel, time range, and task status.
v7.2
Released on March 30, 2026
New Features
Included in this release:
-
Keywords functionality
The transcription API now supports configuring
keywordsto improve the recognition accuracy of specific words such as proper nouns and industry terms. You can configure up to 500 keywords per request. To specify keywords, pass the list in thekeywordsparameter of the request body when creating a transcription task or real-time transcription service. -
Unique
sentence_idWhen parsing transcription data, each subtitle carries a unique identifier
sentence_id. When both original and translated subtitles are enabled, use this ID to match original and translated subtitles for the same sentence, ensuring accurate subtitle alignment.
v7.0
Released on May 1, 2025
Improvements
This release includes the following enhancements:
- Supports updating the configuration of the STT agent.
- Enhances the scalability of STT by supporting multiple endpoints.
- API deprecations. Refer to the API reference for details.
