Azure/azure-sdk-for-net Azure.AI.VoiceLive_1.1.0

on GitHub

1.1.0 (2026-06-03)

Features Added

• Added support for built-in tool calling scenarios, including file search and web search.
• Added lifecycle status tracking for tool calls, including Searching, InProgress, and Completed states.
• Added streaming server events for tool execution, transcript annotations, and video output updates.
• Added avatar voice synchronization support via AzureAvatarSyncVoice with configurable voice parameters.
• Added phrase-level transcription support, including word-level timing, confidence, and phrase grouping.
• Added new personal voice models: PersonalVoiceModels.MaiVoice1 and PersonalVoiceModels.DragonHDOmniLatestNeural.
• Added new turn-detection options, including semantic VAD variants (AzureSemanticVadTurnDetection, AzureSemanticVadTurnDetectionEn, and AzureSemanticVadTurnDetectionMultilingual).

Breaking Changes

• Updated transcription completion APIs to include additional metadata (logprobs and phrases).
• Updated OutputTokenDetails constructors to support reasoning tokens.
• Updated VoiceLiveSessionOptions to include new properties such as Include and Metadata.
• Renamed MCPApprovalType to McpApprovalKind.
• Renamed AvatarConfiguration.Type to AvatarConfiguration.AvatarKind.
• Renamed CustomLexiconUrl to CustomLexiconUri and CustomTextNormalizationUrl to CustomTextNormalizationUri on AzureAvatarSyncVoice, AzurePersonalVoice, AzureCustomVoice, and AzureStandardVoice. The property type changed from string to System.Uri.
• Renamed RequestImageContentPart.Url to RequestImageContentPart.Uri (typed as System.Uri).
• Renamed PhotoAvatarBaseModes to PhotoAvatarBaseMode (singular). AvatarConfiguration.Model was renamed to AvatarConfiguration.BaseMode and now uses the new type.
• Renamed AvatarConfiguration.OutputAuditAudio to AvatarConfiguration.AuditOutputAudio.
• Renamed ReasoningEffort.Xhigh to ReasoningEffort.ExtraHigh.

Bugs Fixed

• Telemetry: gen_ai.event.content on .done and related response events now correctly respects the content recording opt-in (OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT or AZURE_TRACING_GEN_AI_CONTENT_RECORDING_ENABLED). Previously these events emitted message content (transcripts, function-call arguments, response bodies) unconditionally. To restore the previous behavior, set one of those environment variables to true.

Other Changes

• Runnable sample applications moved from sdk/voicelive/Azure.AI.VoiceLive/samples/ to top-level samples/voicelive/. Existing sample folders were renamed to kebab-case (BasicVoiceAssistant → basic-voice-assistant, CustomerServiceBot → customer-service-bot), and each sample's entry point was renamed from SampleProgram.cs to Program.cs.
• Added new samples:
  • samples/voicelive/agent-voice-assistant — Foundry agent integration sample.
  • samples/voicelive/mcp-voice-assistant — MCP server integration sample.
  • samples/voicelive/telemetry-tracing — end-to-end OpenTelemetry tracing and metrics walkthrough for the SDK telemetry shipped in 1.1.0-beta.4.
• Added --show-traces flag and console-tracing wiring to basic-voice-assistant, mcp-voice-assistant, agent-voice-assistant, and customer-service-bot.