The CallEndCoordinator.observe() decremented _tracked_speeches on every
BotStoppedSpeakingFrame, not just the one corresponding to tracked
(end-node) speech. When a preceding LLM utterance finished between
track_speech() and the actual fixed-speech playback, the counter was
consumed prematurely, ending the call before the end-node message played.
Gate the decrement behind a _current_speech_is_tracked flag that is set
only when BotStartedSpeakingFrame follows a pending tracked-speech start.
Co-Authored-By: Claude <noreply@anthropic.com>
- Added support for managing user transcripts, including pending states and timestamps.
- Implemented methods to handle user transcript completion and deferred assistant messages.
- Updated event handling to ensure user transcripts are emitted before assistant responses.
- Enhanced tests to verify the correct order of transcript and assistant message emissions during user interactions.
- Introduced Qwen-Audio Realtime service for speech-to-speech processing in Pipecat.
- Updated interface catalog to include Qwen-Audio Realtime capabilities.
- Enhanced model resource testing to support new service.
- Added configuration options for audio sample rates and turn detection modes.
- Updated documentation to reflect integration details and usage instructions.
- Extend McpTransport to support "sse" in schemas.
- Refactor McpToolClient to handle both "streamable_http" and "sse" transports.
- Introduce McpServerDialog for managing MCP server configurations, including transport settings and tool synchronization.
- Replace McpServersSection with the new dialog component for improved server management.
- Add tests for MCP transport handling and server dialog functionality.
- Introduce WorkflowAgentStage to manage agent stage configurations and enhance interaction with the workflow engine.
- Implement WorkflowEdgeEvaluator for priority-aware edge evaluation, improving routing decisions based on conditions and user turns.
- Update WorkflowBrain to handle user turns and routing more effectively, ensuring agents cannot have only one default path.
- Enhance CallEndCoordinator to track speech events and manage call termination based on queued speech.
- Add new models and output handling for workflow interactions, improving clarity and maintainability.
- Update tests to validate the new routing logic and agent behavior under various scenarios.
- Update WorkflowBrain to handle greeting playback more effectively, ensuring that the initial greeting completes before transitioning to the first node.
- Introduce new methods for managing greeting states and conditions, enhancing the interaction flow for user turns.
- Refactor WorkflowLLMRouter to improve routing logic and ensure proper handling of conditional paths.
- Enhance tests to verify the correct behavior of greeting management and routing under various scenarios, including waiting for audio playback to finish.
- Update frontend components to reflect changes in edge handling and improve user experience in workflow configurations.
- Add begin_response and finish_after_current_speech methods to CallEndCoordinator for better management of speech events.
- Update PromptBrain to utilize new methods, ensuring proper handling of generated closing speech and tool-only calls.
- Enhance tests to verify the correct behavior of speech tracking and response handling in various scenarios, including waiting for audio to finish before ending calls.
- Introduce a new test suite for CallEndCoordinator to validate the interaction with speech frames.
- Introduce greeting context handling in BaseBrain and WorkflowBrain to manage assistant greetings effectively.
- Implement prepare_greeting_context method to add greeting messages to the local context while preserving playback order.
- Update pipeline event handling to ensure greeting timestamps are maintained until the client is ready.
- Enhance tests to verify the correct behavior of greeting context management in various scenarios.
