ShivangiReja

This commit introduces **new capabilities** to the **AudioClient** by adding comprehensive support for the **Voices API**. It enables users to programmatically **create, manage, and retrieve custom voices and voice consents** through new methods like `CreateVoice` and `GetVoiceConsents` in the `AudioClient`. This **feature addition** expands the `AudioClient`'s functionality, allowing for full interaction with the Voices API endpoints across different .NET targets. The change includes new TypeSpec models and operations, along with dedicated test coverage in `VoiceTests.cs` to ensure proper functionality.

This commit introduces **Server-Sent Events (SSE) streaming support for speech generation** within the **Audio** module, providing a **new capability** for real-time audio output. It adds `GenerateSpeechStreaming` and `GenerateSpeechStreamingAsync` methods to the public API surfaces across .NET 10.0, .NET 8.0, and .NET Standard 2.0, enabling incremental delivery of speech data. The implementation includes new streaming-specific data structures like `StreamingSpeechUpdate`, `StreamingSpeechAudioDeltaUpdate`, and `SpeechTokenUsage` to manage the streamed events and their serialization. This enhancement significantly improves the responsiveness of speech generation, allowing applications to process and play audio as it's being synthesized, and is validated with new integration and mock tests.

This commit introduces a **new capability** for **audio transcription** by adding a `ChunkingStrategy` mechanism, allowing users to define how audio is segmented before processing. It provides options for a default chunking strategy and a **custom server-side VAD** (Voice Activity Detection) strategy with configurable parameters like padding and silence duration. This enhancement significantly expands control over the transcription process, affecting **API definitions** across various .NET targets, **TypeSpec models**, and **serialization logic**. The changes enable more flexible and precise audio processing, with new tests validating the functionality of these chunking strategies.

This commit introduces a **new capability** to the **audio transcription API**, enabling **diarized audio transcription** across various .NET client libraries. It adds new public methods like `TranscribeAudioDiarized` and `TranscribeAudioDiarizedAsync`, allowing developers to identify and differentiate speakers within an audio file. This feature is supported by new data models such as `DiarizedAudioTranscription` and `DiarizedTranscriptionSegment`, along with enhanced usage tracking types like `AudioTranscriptionUsage` and `AudioTranscriptionDurationUsage`. Extensive refactoring of existing audio models and the addition of comprehensive tests ensure robust integration and correct functionality for this expanded audio processing feature.

This commit **introduces new integration tests** for the **Realtime MCP tool**, significantly expanding the test coverage for this critical component. The tests, located in `tests/Realtime/RealtimeToolTests.cs`, validate key aspects of the tool's operation, including `tool calls`, `approval policies`, and the management of `allowed tools`. This **enhances the reliability** of the **Realtime** module by ensuring its core functionalities interact correctly and robustly within the system. It represents a **new capability** in terms of testing infrastructure, providing a stronger foundation for future development and maintenance of the `Realtime` tool.

This commit introduces a **new capability** to the **API models and client**, enabling the `Instructions` property within `ResponseResult` and `CreateResponseOptions` to accept either a simple string or a structured `IList<ResponseItem>`. This **enhancement** provides greater flexibility for defining and processing response instructions, moving beyond a single string format. Implementing this required significant **refactoring** of the **TypeSpec definition**, updating **.NET API client generation** for various frameworks, and adding **custom serialization/deserialization logic** to handle the polymorphic type. Consequently, developers can now leverage more complex, structured instructions when interacting with the API. Additionally, several internal tool call parameters were updated from read-only to read-write.

This commit **introduces comprehensive documentation** for the new **`Spec-ingestion` agent skill**, specifically designed to support the **TypeSpec ingestion workflow**. It provides a detailed **new capability** in the form of guides, checklists, and best practices, covering everything from an overview and step-by-step instructions to common patterns, gotchas, and reference materials. This documentation, located under `.github/skills/Spec-ingestion/`, aims to standardize and streamline the process of ingesting TypeSpec specifications, thereby improving efficiency and consistency for future operations.

Add vector store creation example with manual polling (#682)

This commit **refactors** the **Vector Store APIs** to transition them from **Long-Running Operations (LROs)** to **immediate, non-LRO operations**. Specifically, methods such as `AddFileToVectorStore`, `AddFileBatchToVectorStore`, and `CreateVectorStore` in the **OpenAI client library** now return direct results, eliminating the need for LRO polling. This **API change** simplifies the interaction model for managing vector stores and their files, impacting the **public API surface**, **client implementation**, and **TypeSpec specification**. Corresponding **tests** have been updated to reflect this new non-LRO behavior and ensure correctness.

This commit **introduces specific derived types** for `ResponseMessageAnnotation` within the **OpenAI API client**, adding `FileCitationMessageAnnotation`, `FilePathMessageAnnotation`, and `UriCitationMessageAnnotation`. It **refactors** the base `ResponseMessageAnnotation` to remove direct properties, making it a more abstract base for these new, concrete annotation kinds. This **feature addition and refactoring** significantly enhances the type safety and clarity of message annotations, allowing for more precise handling of different citation and file path information. The changes impact the public API definitions, internal model serialization logic, and the `OpenAIModelFactory`, providing consumers with more granular and semantically rich annotation objects.

This commit **enhances the public API surface** of the **OpenAI client library** by making the `GetVectorStoreAsync` method publicly accessible. Previously `internal`, this asynchronous method within the `VectorStoreClient` is now exposed through the `IOpenAIClient` interface for both .NET 8.0 and .NET Standard 2.0 targets. This **API surface enhancement** ensures **consistency** with its synchronous counterpart, aligning the public interface across different .NET frameworks. As a result, external consumers can now directly retrieve vector store information, **improving the usability and completeness of the Vector Stores API**.

This commit **fixes a .NET SDK version mismatch** within the **`update-generator` GitHub Actions workflow**. It updates the workflow configuration in `.github/workflows/update-generator.yml` to use the `global-json-file` option for determining the .NET SDK version, replacing a hardcoded value. This **maintenance change** ensures that the workflow automatically aligns with the project's specified SDK version, preventing build failures caused by version discrepancies. The change improves the **reliability and consistency of the CI/CD pipeline** for generator updates by making the SDK versioning more dynamic and robust.

This commit **fixes a path error** within the **`update-generator` GitHub Actions workflow**. Specifically, it corrects the reference to `codegen/package.json` in the `node -p` command used by the `check-updates` job. This **maintenance fix** ensures that the automated dependency update checks for the `codegen` module can execute successfully, preventing failures in the continuous integration pipeline.

This commit introduces a **new automation capability** to streamline the maintenance of the **TypeSpec generator version**. A new **GitHub Actions workflow** (`.github/workflows/update-generator.yml`) is added to automatically check for and trigger updates on a schedule or manually. This workflow executes a new PowerShell script (`scripts/Submit-GeneratorUpdatePr.ps1`) which handles updating the generator package, modifying project files, regenerating code, and creating a pull request. This **maintenance feature** significantly **reduces manual effort** and ensures the project consistently uses the latest TypeSpec generator, improving dependency management.

This commit introduces a **new capability** to the **ChatClient** by adding `UpdateChatCompletion` and `UpdateChatCompletionAsync` methods. These methods enable users to **update the metadata of previously stored chat completions** through the API. The change involves extending the `ChatClient` interface across different .NET targets, defining new internal request models like `InternalUpdateChatCompletionRequest` with associated serialization logic, and adding a new `ChatCompletionMessageCollectionOrder` struct. This enhancement provides developers with greater control over managing chat completion data, supported by new test cases to verify functionality.

This commit **introduces a new capability** to configure **semantic voice activity detection (VAD)** within the **Realtime API**, allowing users to specify different levels of VAD eagerness. It adds the `SemanticEagernessLevel` struct and a factory method `CreateSemanticVoiceActivityTurnDetectionOptions` to `TurnDetectionOptions`, exposing these new options in the **public API definitions** for both .NET 8.0 and .NET Standard 2.0. The work includes implementing **serialization and deserialization support** for these new semantic VAD options and updating internal models to utilize the new type. This enhancement provides more granular control over voice activity detection behavior, with new smoke tests ensuring proper functionality.

This commit **fixes a serialization bug** within the **Assistants API's `MessageContent` generation** for image content. It corrects the casing and ensures the `detail` parameter, such as `MessageImageDetail.Auto`, is properly serialized using `ToSerialString()` in factory methods like `FromImageFile` and `FromImageUri`. This resolves an issue where image detail settings might not have been correctly transmitted to the API. A new **test case** `ThreadWithImageDetailWorks` was added to `AssistantsTests.cs` to validate the correct handling of image detail in thread creation, ensuring robust functionality for image-based messages.

This commit introduces a **validation mechanism** within the **Audio client** to prevent the use of unsupported models, specifically `whisper-1`, for streaming audio transcription. A new private method, `EnsureModelSupportsStreaming`, is integrated into existing streaming methods like `TranscribeAudioStreamingAsync` and `TranscribeAudioStreaming` to throw an exception if an incompatible model is selected. This **validation enhancement** ensures correct API usage and prevents runtime errors when attempting unsupported operations. The change improves the robustness of the **audio transcription functionality**, with a new test case confirming that the appropriate exception is thrown for the `whisper-1` model.