Independent sound. Human-centered technology.Studio / Software / Research

Community

Conversations
in AI audio.

The tools are changing fast. Explore the questions shaping music technology, rights, ethics, and creative practice.

Verified members can like and join the discussion.

Create an account or sign in

Commercial conversations

Where products meet rights.

Academic conversations

Where ideas meet practice.

Follow the questions

Read further.

Short context and the original sources behind each conversation.

Official ElevenMusic and SoundCloud integration announcement artwork

Commercial / November 6, 2026

What changes when AI music can go straight to a public profile?

ElevenLabs has added a direct SoundCloud connection for original tracks made in ElevenMusic. After linking accounts, a creator can add a description, genre, and tags, then upload the track to a SoundCloud profile. The announcement describes a shorter publishing path, not a new licensing arrangement or a policy for labeling AI-made tracks. As generation and distribution converge, how will listeners understand a track’s origin, and how will platforms keep discovery useful for artists?

Read the source: ElevenLabs · SoundCloud integration announcement

Comments (0)

Colorful gradient artwork from the ElevenLabs SFX v3 announcement

Commercial / October 29, 2026

Can one prompt build a whole sound scene?

ElevenLabs has released SFX v3 in ElevenCreative and its API. The company says the model follows prompts with multiple sound events more reliably than SFX v2 and improves music, vocals, and some high-frequency sounds. That could reduce the number of separate generations needed to sketch a scene, but the performance claims come from the provider and will vary by prompt. The tool remains a generation aid; selecting, timing, and mixing those sounds are still creative decisions.

Read the source: ElevenLabs · SFX v3 announcement

Comments (0)

TYGER's white game-controller mascot

Commercial / October 22, 2026

Can licensed songs become responsive game soundtracks?

UMG and TYGER announced a deal to bring selected catalog recordings to TYGER’s gaming service. TYGER says its music adapts to gameplay without requiring game-studio integration. The UMG catalog is planned for a later rollout; the platform is currently in closed beta. The deal is a concrete commercial test of licensed adaptive music, while the announcement leaves open how artists approve or constrain individual adaptations.

Read the source: Universal Music Group · TYGER licensing announcement

Comments (0)

ElevenCreative artwork announcing Instant Soundtracks

Commercial / October 15, 2026

When a video gets an instant soundtrack, who directs the sound?

ElevenLabs’ Instant Soundtracks scores uploaded video and generates effects aligned to moments it identifies. The company says the effects sit on separate editable layers, and its API can export music and effects separately. Automated timing may accelerate a first pass, but the editor still decides which cues belong, which should move, and how the result fits the picture.

Read the source: ElevenLabs · Introducing Instant Soundtracks

Comments (0)

Excerpt from the U.S. Copyright Office notice of inquiry on music streaming fraud

Commercial / October 7, 2026

Can AI-generated music scale streaming fraud faster than platforms can detect it?

On October 7, the U.S. Copyright Office opened a public inquiry into music streaming fraud. Its notice asks for data on the relationship between AI-generated music, fraudulent plays, and royalty distribution, along with evidence on detection and remedies. It has not reached findings; initial public comments are due November 23.

Read the source: U.S. Copyright Office · Music Streaming Fraud notice of inquiry

Comments (0)

First page of the SPDMX open music stems preprint

Academic / November 6, 2026

Can open music stems ease the training-data bottleneck?

Long and colleagues present SPDMX, a corpus built by rendering public-domain PDMX scores into aligned audio stems. The preprint reports 253,982 songs, 490,159 mono stems, and 5,469 song-hours, released under CC BY 4.0 with scores and an open rendering pipeline. The scale may help research on separation, transcription, and generation without redistributing copyrighted recordings. Because the audio is synthesized from scores, results trained on it still need testing against real performances and production styles.

Read the source: Long et al. · SPDMX preprint

Comments (0)

First page of the game-music theme similarity preprint

Academic / November 3, 2026

How close should new game music stay to its theme?

Luo and colleagues fine-tune a MIDI generation model on paired sections from Super Mario pieces, labeling how similar each section is to its source. Across 157 test contexts, requests for high similarity yielded more closely related outputs than requests for low similarity, especially through melodic restatement. The listening test also found that ratings tracked melodic overlap. This shows a possible control for repeated game music, though the study is limited to one source repertoire and symbolic output; it does not establish how the approach works in a live game or with audio generation.

Read the source: Luo et al. · arXiv preprint

Comments (0)

First page of the CrowdForge audience-participation preprint

Academic / November 2, 2026

When an audience shapes AI music, can they hear their own influence?

Hopkins and colleagues describe a mixed-reality concert with a guitarist and DJ in which the audience collectively influenced AI-generated music and visuals. Their study compared audio, visual, and combined participation. The authors report that visible feedback helped people recognize their influence, while musical controls shaped larger structural changes and sometimes pushed performers into unexpected choices. The study offers design clues for participatory performance, but its findings come from one concert setting and do not establish a universal interface rule.

Read the source: Hopkins et al. · CrowdForge preprint

Comments (0)

First page of the preprint on structured captions for text-to-music generation

Academic / October 31, 2026

Can better descriptions make AI music easier to steer?

Wu and colleagues compare structured captions with free-form text while fine-tuning a text-to-music model. They report better alignment with musical descriptions and improved output quality when the model is trained with structured captions. Their proposed Music DSG measure breaks a prompt into interpretable musical claims and, in their evaluation, correlates with listener preference better than CLAP. The findings are from a preprint submitted to ICASSP 2027; they do not yet establish how well the approach transfers to other models or catalogs.

Read the source: Wu et al. · arXiv preprint

Comments (0)

First page of the SMEBench preprint

Academic / October 27, 2026

What makes an AI music edit successful: the change, or what stays the same?

The SMEBench preprint constructs paired source and target audio from controlled MIDI changes, spanning twelve kinds of edits in instrumentation, articulation, dynamics, and pitch. Its balanced 3,600-example test set and smaller real-audio complement let researchers assess whether an edit happened and whether untouched content survived. In tests of six editing systems, the chosen metric materially changed their ranking; a single score cannot capture every part of a useful edit.

Read the source: Cervera et al. · SMEBench preprint

Comments (0)

Verified members can react and post. Clear responses appear immediately; others may need review. Read the Community Guidelines