Refactor the audio preparation workflow to utilize torchaudio for resampling, replacing librosa for better performance. Introduce a caching mechanism for resampling transforms and update the PrepareRefSemanticBatchWorker to include detailed timing metrics for profiling. Additionally, implement a new CPU limiter for managing resource allocation during audio processing. These changes improve the efficiency and maintainability of the TTS system.
Add resampling capabilities using torchaudio to prepare reference audio at 16kHz, replacing librosa for better performance. Introduce a caching mechanism for resampling transforms to optimize resource usage. Update batch processing methods to include timing metrics for profiling, enhancing the ability to monitor and improve performance in the TTS system. This update improves the maintainability and efficiency of audio preparation workflows.
Add support for G2PW processing in the TTS system by introducing new methods and classes for handling G2PW segments. Update PrepareCoordinator to manage G2PW worker threads and integrate G2PW profiling into the existing framework. Enhance text preprocessing to identify segments requiring G2PW and streamline the resolution of these segments. This update improves the overall performance and maintainability of the TTS system by optimizing the handling of Chinese text processing.
Introduce a new submodule for g2pw and implement AsyncStageGate in PrepareCoordinator to manage concurrent task inflight limits. Update PrepareTextCpuWorker and PrepareRefSemanticBatchWorker to support asynchronous task submission and completion notifications. Enhance profiling capabilities in TTS to track g2pw processing times, improving overall performance and maintainability of the TTS system.
Refactor api_v2.py and api_v3.py to update sampling parameters and weight paths for better clarity and support for v3/v4 vocoders. Introduce new methods in PrepareCoordinator for handling empty text features and improve profiling capabilities. Additionally, update unified engine components to streamline audio processing and state management, enhancing overall performance and maintainability of the TTS system.
Introduce new modules including EngineDecodeStageMixin, EngineDispatchStageMixin, EngineFinalizeStageMixin, EnginePrepareStageMixin, and EngineStageFutureMixin. These components enhance the TTS framework by providing structured methods for managing engine stages, including decoding, dispatching, finalizing, and preparing tasks. The new architecture supports improved state management and asynchronous operations, significantly enhancing the maintainability and performance of the TTS system.
Introduce new modules including unified_engine_component_models, unified_engine_component_policy, unified_engine_component_registry, unified_engine_component_runtime, unified_engine_worker_completion, and unified_engine_worker_decode. These additions enhance the TTS framework by providing structured models for request handling, engine policies, and worker execution, significantly improving the architecture and maintainability of the system. The new components support asynchronous operations and optimize overall performance through better state management and processing capabilities.
Introduce new modules including unified_engine_api_direct, unified_engine_api_profile, unified_engine_api_request, and unified_engine_api_scheduler. These additions enhance the TTS system by providing structured interfaces for direct TTS execution and scheduler-based processing. The new components support improved request handling, profiling, and state management, significantly enhancing the architecture and maintainability of the TTS framework.
Introduce new modules including EngineApiDelegates, EngineBridgeDelegates, EngineRegistryBridgeFacade, EngineRuntimeBridgeFacade, EngineStageBridgeFacade, and EngineStageOrchestrator. These additions provide a structured approach to managing TTS requests, engine states, and orchestration, significantly improving the architecture and maintainability of the TTS system. The new components support asynchronous operations and enhance overall performance through better request handling and processing capabilities.
Rename several methods in EngineApiFacade to follow a consistent private naming convention, enhancing code clarity. Update EngineApiDelegates to remove redundant method definitions, streamlining the interface. Introduce EnginePublicInterface to encapsulate public API methods, improving organization and maintainability of the TTS system. Additionally, update the EngineCompositionBuilder to use the new scheduler worker state retrieval method.
Introduce multiple new modules including unified_engine_api, unified_engine_audio, unified_engine_bridge, unified_engine_builder, unified_engine_components, unified_engine_delegates, and unified_engine_runtime. These additions provide a comprehensive framework for managing TTS requests, audio packing, and engine state management, significantly improving the architecture and maintainability of the TTS system. The new structure supports asynchronous operations and enhances overall performance through better request handling and processing capabilities.
Added thread lock to protect get_phones_and_bert method from potential race conditions during concurrent access. This addresses issue #1844 where multiple threads accessing the BERT model simultaneously could cause data inconsistency or crashes.
Co-authored-by: RVC-Boss <129054828+RVC-Boss@users.noreply.github.com>
* add exception handling
* Fill in the missing content
* remove GB code
* Simplify i18n text and remove trailing spaces
* ignore v3 model dir
* Update Changelog
* Fix encoding
