MusicTwins post-trains Audio Flamingo Next to match a listener persona
Original title: MusicTwins: Training Audio-Language Models to Listen like Their Persona
On ten held-out personas, fine-tuning and reinforcement learning improved personality, style, and music responses.
MusicTwins is a framework for person-specific responses to text and music from audio-language models. Persona summaries built from human writings capture background, views, personality, writing style, and a musical profile, then combine with personal memory and post-training. The authors post-train Audio Flamingo Next with supervised fine-tuning and reinforcement learning. Evaluation on ten held-out personas reports notable gains in persona fidelity, including responses to music.
- The work targets whether audio-language models can reproduce one person's reactions.
- Persona data comes from human writings rather than a generic listener profile.
- Post-training uses supervised fine-tuning and reinforcement learning.
- Held-out evaluation covers personality, writing style, and music responses.
- AI models
- Audio Flamingo Next
From the original article
1 min read at arxiv.orgPeople differ in how they perceive music, what associations they make with it, what they like about it, and how they respond to it. Computational models intended to simulate people in environments that include music should account for these differences.
By Milan Liessens Dujardin · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.