Azure/azure-sdk-for-java com.azure+azure-ai-voicelive_1.0.0

on GitHub

1.0.0 (2026-06-01)

This is the first General Availability (GA) release of the Azure VoiceLive client library for Java.

Breaking Changes

• Narrowed VoiceLiveAsyncClient session startup to three overloads:
  • startSession()
  • startSession(String, VoiceLiveRequestOptions)
  • startSession(AgentSessionConfig, VoiceLiveRequestOptions)
• Renamed token-count accessors on token statistic models (JSON wire format unchanged):
  • CachedTokenDetails.getTextTokens() / getAudioTokens() / getImageTokens() → getTextTokenCount() / getAudioTokenCount() / getImageTokenCount()
  • InputTokenDetails.getCachedTokens() / getTextTokens() / getAudioTokens() / getImageTokens() → getCachedTokenCount() / getTextTokenCount() / getAudioTokenCount() / getImageTokenCount()
  • OutputTokenDetails.getTextTokens() / getAudioTokens() / getReasoningTokens() → getTextTokenCount() / getAudioTokenCount() / getReasoningTokenCount()
  • ResponseTokenStatistics.getTotalTokens() / getInputTokens() / getOutputTokens() → getTotalTokenCount() / getInputTokenCount() / getOutputTokenCount()
• RequestImageContentPart URL accessor renamed and JSON field changed:
  • getUrl() / setUrl(String) → getImageUrl() / setImageUrl(String)
  • JSON property url → image_url
• Renamed base event types for client↔server symmetry:
  • ClientEvent (base for outbound events) → SessionClientEvent
  • SessionUpdate (base for inbound events) → SessionServerEvent
  • VoiceLiveSessionAsyncClient.receiveEvents() now returns Flux<SessionServerEvent>
  • VoiceLiveSessionAsyncClient.sendEvent(...) now accepts SessionClientEvent
• Renamed MCP-related model types to Pascal case (MCP* → Mcp*): McpApprovalType, McpServer, McpTool, McpApprovalResponseRequestItem, ResponseMcpApprovalRequestItem, ResponseMcpApprovalResponseItem, ResponseMcpCallItem, ResponseMcpListToolItem.
• VoiceLiveSessionAsyncClient.truncateConversation(String, int, int) now accepts a java.time.Duration for the audio-end position instead of raw milliseconds. The two-argument overload (itemId, contentIndex) is preserved and defaults to Duration.ZERO.
• Removed sendInputAudio(byte[]); use sendInputAudio(BinaryData) (wrap raw bytes with BinaryData.fromBytes(...)).
• AgentSessionConfig.toQueryParameters() is no longer part of the public API; the conversion is handled internally by VoiceLiveAsyncClient.
• VoiceLiveSessionOptions.setAnimation(...) renamed to setAnimationOptions(...).
• AnimationOptions.setOutputs(...) / getOutputs() renamed to setOutputTypes(...) / getOutputTypes().
• LogProbProperties.getLogprob() renamed to getLogProb().
• SessionUpdateConversationItemInputAudioTranscriptionCompleted.getLogprobs() renamed to getLogProbs().
• Removed preview service versions from VoiceLiveServiceVersion; only GA versions remain (V2025_10_01, V2026_04_10). The latest version is now V2026_04_10.

Features Added

• Avatar voice synchronization for video avatars:
  • New AzureVoiceType.AVATAR_VOICE_SYNC and AzureAvatarVoiceSyncVoice class
  • New server events ServerEventSessionAvatarSwitchToSpeaking / ServerEventSessionAvatarSwitchToIdle
  • New ServerEventResponseVideoDelta for streaming avatar video frames
  • New ClientEventOutputAudioBufferClear (output_audio_buffer.clear) and ServerEventOutputAudioBufferCleared (output_audio_buffer.cleared) for clearing the avatar output audio buffer
• Web search and file search tool calls:
  • New ItemType.WEB_SEARCH_CALL, ItemType.FILE_SEARCH_CALL
  • New ResponseWebSearchCallItem (with ResponseWebSearchCallItemStatus) and ResponseFileSearchCallItem (with ResponseFileSearchCallItemStatus, plus FileSearchResult results)
  • New lifecycle server events: ServerEventResponseWebSearchCall{Searching,InProgress,Completed} and ServerEventResponseFileSearchCall{Searching,InProgress,Completed}
• Transcription enhancements:
  • New transcription models on AudioInputTranscriptionOptionsModel: GPT_4O_TRANSCRIBE_DIARIZE, MAI_TRANSCRIBE_1
  • New TranscriptionPhrase and TranscriptionWord types with timing/confidence information
  • SessionUpdateConversationItemInputAudioTranscriptionCompleted now exposes getLogProbs() and getPhrases()
  • New ServerEventResponseAudioTranscriptAnnotationAdded event
• Session include options and metadata:
  • New SessionIncludeOption expandable enum for opting into additional response payloads (e.g. logprobs, phrases, file-search results)
  • VoiceLiveSessionOptions and VoiceLiveSessionResponse now expose include (List<SessionIncludeOption>) and metadata (Map<String,String>, up to 16 entries)
• Personal voice models: added PersonalVoiceModels.DRAGON_HDOMNI_LATEST_NEURAL and MAI_VOICE_1
• Reasoning token usage: OutputTokenDetails.getReasoningTokenCount() exposes reasoning token counts
• Interim response on response.create: ResponseCreateParams.setInterimResponse(BinaryData) lets callers attach interim response config to a single response request
• Restored no-arg VoiceLiveAsyncClient.startSession() overload (uses the deployment's default model).
• Significantly improved Javadoc for ServerVadTurnDetection, AzureCustomVoice, AzurePersonalVoice, AzureStandardVoice, AzureSemanticVadTurnDetection*, and other model types

Other Changes

• Updated default service API version to 2026-04-10 (GA).