Amazon and ethics of genAI for music
Summary of Amazon as an uncertified company working on generative AI for music. Includes some product announcements, key features, and a few insights on ethics of AI for music.
Our PART 3 article provides the big-picture view of 85 companies offering AI-based tools or platforms in this musical melody space. Our PART 3 article provides the big-picture view of 85 companies offering AI-based tools or platforms in this musical melody space. This profile page is meant to provide insights into why the company is significant enough to review, even though it wasn’t chosen for deeper dives.
Acronyms
DAW - digital audio workstation
TTA - text to audio (more general than music; can include spoken vocals as well as sung vocals & instruments)
TTM - text to music
Amazon Music
Description from site: “Amazon Music reimagines music listening by enabling customers to unlock millions of songs, podcasts, and curated playlists. Amazon Music provides unlimited access to new releases and classic hits. Our premium subscription service, Amazon Music Unlimited, provides access to more than 60 million songs. Engaging with music and culture has never been more natural, simple, and fun.”
Description from PitchBook: “Amazon is the leading online retailer and marketplace for third party sellers. Retail related revenue represents approximately 75% of total, followed by Amazon Web Services' cloud computing, storage, database, and other offerings (15%), advertising services (5% to 10%), and other the remainder. International segments constitute 25% to 30% of Amazon's non-AWS sales, led by Germany, the United Kingdom, and Japan.”
History and Partnerships
Amazon was founded in 1994 and is headquartered in Seattle, WA. They are publicly traded as AMZN on NASDAQ with a current market cap around $2T. They have about 1.5m employees worldwide.
Amazon Music was launched in public beta in Sept. 2007 and quickly rolled out worldwide. Amazon was the first music store to sell songs without DRM from the 4 major music labels (EMI, Sony BMG, Universal, and Warner) as well as independent labels. Their current catalog is estimated by outsiders to be over 100 million songs (ref). However, as of July 18, 2024, Amazon’s own site says Amazon Music Unlimited users have access to 60 million.
It’s worth noting that Amazon’s business interests in use of AI to generate voices, audio, or music likely extend well beyond “Amazon Music”. Their other entertainment subsidiaries include Amazon Originals (movies), Prime Video, Audible (podcasts and audio books), Amazon Games, Twitch, Amazon Music, and Prime Gaming. All of those lines of business could take advantage of AI-generated background music or voice clones, including translations of dialogue to other languages. And they are already using generative AI in their Ads division (ref) and have built a ML-based music recommendation system for integrating Alexa with Amazon Music (ref).
So even if Amazon isn’t offering (selling) genAI-based tools for sound and music to end users, they could profit greatly from building and using them internally.
In addition to their own Amazon Originals movies, Amazon owns a large library of assets which they can use for training AI. The MGM acquisition in May 2021 for $8.45b gave them 4000+ movies and 17,000 episodes of TV programming; and that’s not all they own or have licensed. They spend billions each year to license content to show on Prime Video.
I’m wondering: do those movie studio licensing agreements allow (or disallow) Amazon to train their own AI models on the content?
Timeline of major events
Dec. 2019: Amazon bundled a MIDI keyboard with a developer kit in their initial Deep Composer offering (site, repo, developer guide, link1, link2, link3)
Feb. 20, 2023: Amazon Music announced a “playlist partnership” with generative AI company Endel, a “Berlin-based AI sound wellness app” (site, link1, link2, link3, link4), aimed at creating music for therapeutic or wellness purposes.
A note on diversity and biases: Amazon announced on June 11, 2024 that their head of DEI for MGM Studios and Prime Video is leaving the company.
Amazon partner Endel has been certified as Fairly Trained. As of July 25, 2024, no Amazon business unit has been certified to an AI ethics standard.
Key Features
Deep Composer was positioned as a tool for developers - not for musicians or non-developers who want to create music. Its goal is to allow developers to “use the keyboard to create a melody that will transform into a completely original song in seconds, all powered by AI”. Its offering bundled a not-quite-required proprietary MIDI keyboard with a developer kit, with pre-trained models for “rock, pop, jazz, and classical” and support through Sagemaker for custom genres.
Although their developer guide shows a 2024 copyright date, it’s not clear what (if anything) has changed since 2019. The GitHub repo appears to be inactive.
Training Data and Technologies
A little information is available about Amazon’s music-related work in areas outside of generating musical melodies. Example: music recommender for Alexa. (“The Amazon Music conversational recommender is hitting the right notes”, by Sean O'Neill, 2022-01-28.)
No information is available about the data used in their prior work on Deep Composer.
What’s Next?
Amazon seems to have shifted their external focus on generative AI for music to partnerships such as with Endel (and likely continuing internal R&D).
Because of the initial developer focus and the apparent shift to partnerships, we will not discuss Amazon further in this article series.
This post is one of the last big pieces in PART 3. PART 4 is coming up next. Subscribe for FREE to be notified automatically of new posts and support our work:
REFERENCES
See this “AI for Music” page for links to all posts related to this series, as well as bonus articles mentioned above on voice cloning and other music-related topics.
