Core Insights - Kunlun Wanwei officially launched the Mureka V7.5 model on August 15, enhancing the performance of Chinese song interpretation significantly [2] - The Mureka V7.5 model demonstrates a deep understanding of various Chinese music styles, allowing for accurate emotional and artistic expression in generated music [2] - The company also introduced MoE-TTS, a novel speech synthesis framework that combines pre-trained large language model capabilities with specialized speech expert modules [3] Group 1 - Mureka V7.5 has improved the timbre and performance techniques of Chinese songs, as well as the articulation and emotional expression [2] - The model's deep accumulation of knowledge regarding Chinese music diversity enables it to convey unique artistic essence and emotional nuances [2] - The ASR technology has been optimized to enhance the authenticity and emotional depth of vocal performances in generated music [2] Group 2 - MoE-TTS innovatively integrates pre-trained large language model text capabilities with speech expert modules, ensuring independent optimization of each modality [3] - The release of MoE-TTS provides a reproducible open descriptive TTS solution for academia and demonstrates the potential of decoupled modalities and knowledge freezing in speech synthesis [3] - Future plans for MoE-TTS include integration into the Mureka-Speech platform, offering customizable descriptive speech synthesis capabilities for global developers and creators [3]