Domestic AI Music Model Mureka V7 Released: Technical Upgrades and Functionality Measurements

With the continuous evolution of AI technology, the generation quality and popularity of AI music is rapidly increasing. The obvious shortcomings of early AI music in terms of sound fidelity and vocal naturalness are being improved with the rapid iteration of the model.

On July 23, 2025, Quintessence officially unveiled its next-generation music mega-model Mureka V7. According to official information, the model outperforms similar products overseas in a number of key indicators such as average performance scores, mix quality, vocal realism and overall sound quality Suno (used form a nominal expression) V4.5 Version. As with its predecessor V6 Compare.Mureka V7 Significant improvements in the richness of melodic motifs and the quality of arrangements have been realized, while enhancing the realism of vocals and instruments.

Mureka V7 has been fully launched on its official website for users to experience.

Functionality in action: from tone imitation to style reference

Mureka V7 One of its core features is "Custom Singers". This feature allows users to upload audio or provide a link to a video, allowing the AI model to learn and mimic a specific tone to sing a brand new song.

Take singer Faye Wong's timbre as an example, her vocals have a unique airy sound and airy sound processing skills. In the use of Mureka V7 When its timbre was imitated and reinterpreted in Qingpingtiao, the generated results reproduced the vocal qualities of the original singer to a large extent, especially in the treatment of the end notes, which embodied a similar sense of lethargy.

In addition, "Music Reference" is another useful feature. The model analyzes the music uploaded by users, identifies its style, rhythm, orchestration and mood, and generates original compositions with similar styles. For example, the recently popular social media song "Just Bought a Plane and Got Hit", adapted from the Indian song "Tunak Tunak Tun", is used as a reference.Mureka V7 Ability to generate tracks with similar melodies and rhythms that are automatically paired with stylized visual music videos.

In terms of general functions, the model supports generating different styles of music directly from text descriptions. Inputting Li Bai's poem "Will Enter the Wine" and specifying the style of "Rap Metal", the model can generate a song that combines the poem with rock elements. For background music (BGM) creation, users can generate copyright-free pure music clips with simple prompts (e.g., "recalling the warm piano melody of childhood") or upload reference audio (e.g., the theme song of "Summer" or "Game of Thrones") to create similar styles of music.

In the case of unsatisfactory generation of resultsMureka V7 Provides basic audio editing tools with support for local modification, song extension, instrument splitting and audio cropping, and is compatible with music creation in ten languages.

Technology core: the evolutionary MusiCoT chain of thought

Mureka V7 performance improvement thanks to its self-developed music generation-specific thought chain MusiCoT (Analyzable Chain-of-Musical-Thought Prompting) for continuous optimization.

In the field of large-scale language modeling, Chain-of-Thought (CoT) is a cueing method that guides a model through a step-by-step reasoning process before answering a question, in order to improve the accuracy of complex tasks. The core logic of CoT is to "think about the structure before generating", which simulates the creative process of human musicians. In the output of specific audio Token Beforehand, the model preplans the overall structure of the music, including passages, emotional progression, and choreographic layout.

MusiCoT Another feature of the generation structure is its interpretability and controllability. With the help of CLAP (contrastive language-audio pre-training model), the chain of thought in which AI generates music becomes explicitly readable. This allows the user to more precisely control the generated result by inputting a reference audio of any length as a stylistic cue. Compared to Suno and other models in the exploration of musical structure and controllability.Mureka (used form a nominal expression) MusiCoT A more interpretable technical path is provided.

A New Model for Speech Synthesis: Mureka TTS V1

In addition to music generation, Kunlun also released a new audio model this time around Mureka TTS V1, specializing in general-purpose speech synthesis.

Unlike music models that emphasize melody and harmony, TTS (Text-to-Speech) models are more concerned with the generic representation of various types of sounds.Mureka TTS V1 The main innovation is the introduction of Voice Design Capabilities that allow users to define the characteristics of a desired timbre through natural language text descriptions, rather than being limited to selecting from a library of preset tones. Users can describe the gender, age, emotion, intonation style and speech rate of the voice, enabling highly personalized speech synthesis.

According to the official published comparison data, in comparison with the industry's mainstream competitors ElevenLabs TTS V2 In a comparison of theMureka TTS V1 It possesses strengths in speech quality, naturalness of subjunctive and utterance rhythm, and overall auditory perception, but falls slightly short in pronunciation accuracy. This suggests that theMureka TTS V1 It differentiates itself in terms of "creativity" and "definability" of sound, and is especially suitable for film, TV, games, advertisements and other scenarios that require a high degree of customization of voice-overs.

For example, if you input "a female child voice, about 12 years old, with a clear and pleasant voice, full of enthusiasm" or "a male newsreader, with a clear and stable voice, calm and rational tone", the model will be able to generate audio clips that match the description, realizing the whole process of generating audio clips from the creative description to the sound output. The model can generate audio clips that match the description, realizing the whole process of generation from creative description to sound output.

In large models Scaling laws Against the background of weakening marginal benefits of AI, the focus of AI industry is gradually shifting to the landing of applications in vertical fields. Through its continuous investment in AIGC creation fields such as music and audio, Kunlun Wanwei aims to open up the transformation path from the underlying technology to application products and seize the ecological position in the content creation scene. Since its debut in April 2024, theMureka The model has attracted a large number of users through rapid iteration, demonstrating the popularity of its product in the market.