Home / Technology / AI Music Generator Aims for Human-Like Sound
AI Music Generator Aims for Human-Like Sound
21 Aug
Summary
- New AI models focus on song cohesion and natural development.
- AI can now reassess musical choices during song creation.
- Goal is music that resonates on a human level and invites replays.
Mureka has introduced its V9.5 AI music generation model and is previewing the upcoming O3 model, both engineered to enhance creativity and completeness in AI-generated music. The O3 model utilizes "test-time scaling" to allow the AI to reassess and refine musical choices as a song progresses, ensuring a more natural flow and focused emotional arc.
This new approach prioritizes songs that feel cohesive from beginning to end, with ideas developing purposefully and sections connecting organically. Mureka's goal is to create music that appeals to listeners over repeated plays, aiming for a "human" resonance rather than solely immediate impact. The platform currently allows users to generate tracks based on mood, style, or lyrical prompts.
Internal testing of V9.5 and O3 involved 100 prompts and 20 evaluators, focusing on comparable conditions for consistency. V9.5 emphasizes selective arrangement and vocal connection, while O3 builds on this with its "reflective test-time scaling" capability. User feedback has been crucial, shaping the focus on realism, coherence, and replay value.
