{
  "id": "music-streaming",
  "title": "Music streaming",
  "lang": "en",
  "source_lang": "en",
  "summary": "Music streaming delivers recorded music over networks for immediate playback, organizing access through licensed catalogs, subscriptions, advertising, and recommendation systems.",
  "category": "arts",
  "subcategory": "music",
  "type": "concept",
  "aliases": [
    "Streaming music",
    "online music streaming"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/music-streaming/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/music-streaming/",
      "json": "https://aiwiki.page/zh/music-streaming.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Music streaming is the delivery of recorded music through a network so that playback can begin before the entire recording has been transferred to the listener’s device. Most contemporary services provide access to large, searchable catalogs through applications, websites, and connected devices. Unlike purchasing a record or an unrestricted digital file, streaming usually provides continuing access under a service’s licensing agreements and terms of use. Some subscriptions include offline listening, but the downloaded recordings normally remain controlled by the application rather than becoming independently usable files.\n\nStreaming combines a technical method of distribution with a set of commercial and cultural institutions. Its development depended on the expansion of the internet, audio compression, broadband and mobile networks, and systems for licensing recordings and musical compositions. By the 2020s, it had become the largest source of revenue for the global recorded-music industry. According to the International Federation of the Phonographic Industry (IFPI), streaming accounted for 67.3 percent of global recorded-music revenue in 2023, within a market totaling US$28.6 billion.\n\nThe medium has changed how listeners encounter music, how recordings circulate, and how income is allocated. Search, playlists, automated recommendations, and continuous listening coexist with albums, artist catalogs, and radio-like programming. Understanding streaming therefore requires distinguishing its network technology from the rights, payment systems, and editorial decisions that determine what is available and who benefits from its use."
    },
    {
      "heading": "Forms of streaming and the meaning of access",
      "text": "Music streaming encompasses several forms of listening that differ in how much control they give the listener. On-demand services allow users to select particular recordings, repeat them, and assemble listening sequences. Noninteractive services provide programmed channels or personalized stations in which the service largely determines the next recording. Between these are hybrid products offering limited skips, shuffled playback, or different controls according to subscription tier, device, and country. These distinctions are commercially important and can also determine which licenses a service needs.\n\nStreaming is not identical to live broadcasting. A studio recording released decades earlier can be streamed on demand, while a live concert can be transmitted as it happens. Nor does streaming necessarily mean that no data are stored locally. Applications generally buffer several seconds or more of audio and may cache frequently used material. The defining technical feature is that playback proceeds while data are being delivered, rather than requiring completion of an ordinary file download before listening begins.\n\nCommercially, the more important distinction is often between ownership of a copy and conditional access. A purchased compact disc remains playable without an account with the retailer. An unrestricted downloaded audio file can generally be copied between compatible devices, subject to applicable intellectual property law. A streaming subscription instead supplies an entitlement to use a changing service. Offline listening usually depends on periodic authentication, an active subscription, and continuing authorization for the recording. A track can disappear following expiration of a license or a rights dispute, even if a listener previously saved it to a library.\n\nThe term “library” consequently has several meanings. It may describe the service’s entire licensed catalog, a user’s collection of saved references to recordings, or locally stored files uploaded or matched by the service. Saving an album usually creates an entry in an account rather than transferring ownership of its audio. Similarly, following an artist, liking a track, and adding a song to a playlist are distinct actions that can generate different recommendation signals.\n\nThe boundaries of the category also depend on what is being measured. Audio-only subscription services, advertising-supported music videos, internet radio, and music embedded in short-form video all deliver music through networks, but industry reports do not always group them together. Podcasts and audiobooks may share the same application without belonging to its music catalog. Subscriber counts, listening time, advertising revenue, and payments to music rightsholders therefore describe related but nonidentical markets."
    },
    {
      "heading": "Historical development",
      "text": "Streaming emerged from several earlier developments rather than from a single invention. Digital audio made recordings reproducible as computer data, while compression reduced the bandwidth and storage needed to distribute them. Early consumer internet connections were too slow and inconsistent for convenient transmission of high-quality uncompressed audio. During the 1990s, lower-bitrate formats and buffering made continuous network playback practical, although sound quality and reliability remained limited.\n\nRealAudio, introduced by Progressive Networks in 1995, was an important early platform for streaming sound. Internet radio stations and broadcasters used such systems to reach listeners beyond terrestrial transmission areas. These services established a basic streaming experience: a user connected to a remote source, waited for a buffer to fill, and listened while the connection continued. They generally offered less individual control than later on-demand catalogs.\n\nThe growth of MP3 file sharing created a different model of online music circulation. Napster, launched in 1999, helped users locate and exchange recordings through a centralized indexing system. It was principally a file-sharing service rather than a modern licensed streaming platform. Its rapid adoption and subsequent litigation nevertheless demonstrated both demand for convenient digital access and the difficulty of reconciling that demand with existing rights arrangements. Apple’s iTunes Music Store, launched in 2003, developed a successful paid-download alternative in which users bought individual tracks and albums.\n\nLicensed subscription streaming also developed during this period. Rhapsody launched in 2001 with on-demand access, while Pandora’s service, introduced in 2005, emphasized personalized radio. Spotify launched in Sweden in 2008 and expanded the combination of advertising-supported access and paid subscriptions. Deezer began operating in France in 2007. Later entrants included Apple Music in 2015 and YouTube Music in 2015, alongside subscription products from Amazon and other technology companies. These services differed in geographic reach, integration with hardware, free-tier design, and relationships with existing media businesses.\n\nSmartphones and improving mobile data networks were decisive in moving streaming beyond desktop computers. Listeners could carry large searchable catalogs without managing equivalent local storage. Offline modes reduced dependence on continuous mobile reception, while wireless speakers, connected televisions, car systems, and voice interfaces made streaming part of domestic and everyday listening.\n\nThe transition was gradual and geographically uneven. Physical recordings remained important in some markets, and downloads did not disappear when streaming expanded. The change was nevertheless substantial: recorded-music businesses increasingly earned revenue from repeated licensed access rather than primarily from selling copies. Earlier catalogs could generate continuing income, while new releases entered an environment where availability, discovery, and sustained attention mattered alongside initial sales."
    },
    {
      "heading": "Audio delivery, compression, and playback",
      "text": "A streaming service typically begins with an audio master supplied by a record label, distributor, or other authorized provider. The service processes the file into one or more delivery formats, checks technical characteristics, and associates it with catalog information. Playback then involves retrieving the appropriate audio, confirming access rights, transmitting data, decoding it, and sending the resulting signal through the device’s audio system. Several independent technical choices affect the result.\n\nAudio codecs reduce or organize the data needed to represent sound. Lossy codecs, including AAC, MP3, and Ogg Vorbis, discard information according to perceptual models intended to limit audible effects. Lossless formats such as FLAC compress audio without changing the decoded sample values. “Lossless” describes preservation of those values; it does not establish that a recording has a better master, that the listener’s equipment reproduces every difference, or that all stages of a wireless playback chain remain lossless.\n\nBitrate expresses the amount of encoded data transmitted per unit of time. For uncompressed pulse-code modulation, it can be calculated as\n\n\\[R=f_sbn,\\]\n\nwhere \\(f_s\\) is the sampling rate, \\(b\\) the bit depth, and \\(n\\) the number of channels. Stereo audio at 44,100 samples per second and 16 bits per sample requires 1,411,200 bits per second before file overhead. Lossy streaming uses substantially lower rates. At a constant 320 kilobits per second, an hour of audio contains approximately 144 megabytes of encoded data, excluding protocol overhead.\n\nServices use distributed servers and content delivery networks to avoid sending every request from a single location. Popular recordings can be cached closer to listeners, reducing delay and network congestion. The application maintains a buffer so that brief fluctuations in delivery do not interrupt playback. Some systems change quality according to connection conditions; others allow users to choose separate settings for mobile networks, Wi-Fi, and offline storage.\n\nSeveral playback features operate independently of compression. Gapless playback preserves continuity between tracks intended to join without silence, a requirement for many live recordings, operas, and continuous album sequences. Crossfade deliberately overlaps adjacent recordings. Loudness normalization adjusts playback gain to reduce large perceived level differences between tracks; it does not necessarily alter the stored master or compress its internal dynamic range. Multichannel and spatial-audio versions introduce additional requirements for mixing, format support, and rendering.\n\nThe final sound also depends on the recording itself, device settings, headphones or speakers, and the listening environment. A higher advertised bitrate cannot correct distortion in the source master. Conversely, audible interruptions may reflect network congestion or application behavior rather than the recording format. Technical comparisons therefore distinguish the source, encoding, transmission, decoding, and acoustic reproduction stages."
    },
    {
      "heading": "Copyright, licensing, and catalog organization",
      "text": "A commercially released recording ordinarily contains at least two distinct rights layers: the musical work and the particular sound recording. The work includes the composition and, where applicable, lyrics; the recording embodies a specific performance. A songwriter or publisher may control rights in the composition, while a label, artist, or another producer controls the recording. Performing on a recording does not by itself establish ownership of either layer.\n\nStreaming services must obtain authorization appropriate to the rights and territories involved. Copyright systems differ between countries, and the legal treatment of an interactive stream may differ from that of noninteractive radio. In the United States, qualifying noninteractive digital audio transmissions can use a statutory sound-recording license administered through SoundExchange. Interactive on-demand services generally negotiate sound-recording rights directly with rightsholders. Rights in compositions involve separate licensing and payment arrangements.\n\nThe United States’ Music Modernization Act of 2018 changed the administration of certain digital mechanical rights. The Mechanical Licensing Collective began administering the new blanket license for eligible interactive streams and downloads in January 2021. This did not replace all music licensing: public-performance rights in compositions, sound-recording permissions, and rights outside the United States remained separate. In other jurisdictions, collecting societies and other licensing bodies operate under different legal frameworks.\n\nCatalog availability follows these agreements rather than simply the physical location of a server. A recording may be licensed in one country but not another, and different companies may control it in different territories. Releases can also be withdrawn, replaced, or transferred between distributors. Public-domain status is similarly layered: a composition may no longer be protected while a recent recording of it remains protected.\n\nAccurate metadata are essential to identifying what is being used and directing payments. The International Standard Recording Code, or ISRC, identifies a recording, while the International Standard Musical Work Code, or ISWC, identifies a musical work. Product identifiers distinguish releases, and contributor fields identify performers, writers, producers, and other participants. These systems are related but not interchangeable. One composition can have many recordings, and the same recording can appear on an original album, a compilation, and a reissue.\n\nErrors can arise from inconsistent titles, missing songwriter information, duplicate entries, conflicting ownership claims, and incomplete links between recordings and compositions. Such problems can delay allocation of money even when playback has been counted correctly. They also affect search: a movement from a symphony may need to be connected to its larger work, composer, conductor, orchestra, and particular recording. Catalog organization is therefore part of both rights administration and the listener’s ability to find the intended performance."
    },
    {
      "heading": "Business models and royalty allocation",
      "text": "The principal consumer business models are paid subscriptions and advertising-supported access. Subscriptions may be sold to individuals, students, households, or through bundles with telecommunications and other services. Advertising-supported tiers exchange some listening freedom or uninterrupted playback for access without a direct subscription payment. Services may also combine music with other media, making the allocation of subscription revenue more complex than the price of a single-purpose music product suggests.\n\nRevenue received by a service is not identical to revenue paid to artists. Payments first pass through contractual and legal arrangements involving recording owners, distributors, publishers, collecting organizations, and other intermediaries. An artist’s eventual income depends on ownership, royalty rates, recoupment provisions, and any division among collaborators. A performer who owns a recording and uses a distribution service occupies a different contractual position from one recording under a traditional label agreement.\n\nA common allocation method is the pro-rata, or market-share, model. In simplified form, a rightsholder’s allocation from a defined royalty pool can be represented as\n\n\\[P_i=R\\frac{S_i}{\\sum_j S_j},\\]\n\nwhere \\(R\\) is the distributable pool and \\(S_i\\) is the number of eligible streams attributed to rightsholder \\(i\\). Actual arrangements can include different territories, service tiers, contractual guarantees, eligibility rules, and other adjustments. The formula describes a principle rather than a universal contract.\n\nConsequently, there is no single fixed “per-stream rate” applying across the industry. A reported average may be calculated by dividing a payment by the streams associated with it, but that average depends on the underlying revenue and accounting period. Subscription prices, advertising demand, exchange rates, territory, free-tier use, and contractual terms all influence the result. Figures referring to sound-recording payments also cannot automatically be treated as including songwriter income.\n\nAn alternative is user-centric allocation, in which each subscriber’s distributable payment is divided among the recordings that subscriber actually plays. This differs from pooling all eligible listening within a market or tier. The relative effect depends on listening patterns: users vary in both the amount and the kinds of music they consume. Results also depend on which rights are included, how household accounts are treated, and how administrative costs and fraudulent activity are handled. Neither system makes income equal among artists.\n\nStreaming’s commercial importance is evident in industry totals, but recorded-music revenue does not represent the entire music economy. Concert tickets, merchandise, music publishing, synchronization licensing, and other income sources may be accounted for separately. IFPI’s 2023 figures placed subscription streaming at 48.9 percent of global recorded-music revenue, with advertising-supported streaming supplying a further 18.5 percent. These figures measure industry revenue categories, not the share of each listener’s spending that ultimately reaches an individual musician."
    },
    {
      "heading": "Discovery, playlists, and recommendation systems",
      "text": "Streaming catalogs contain more recordings than any listener can meaningfully inspect. Discovery interfaces therefore help determine which part of a catalog becomes visible. Search serves explicit intentions, such as finding an artist or album; recommendations address less specific intentions, such as choosing unfamiliar music or continuing a listening session. Editorial playlists, personalized mixes, charts, artist pages, and automatically generated stations are distinct mechanisms, even when presented together in one application.\n\nA recommender system may combine several kinds of evidence. Collaborative filtering identifies patterns among users and recordings, such as listeners who repeatedly play overlapping sets of artists. Content-based methods use information about the recordings themselves, including genre descriptions, text, and measured audio characteristics. Hybrid systems combine these approaches with context such as recent listening, device, and time of day. The precise combination varies by service and product.\n\nMachine learning can represent users and tracks in numerical spaces where proximity reflects inferred relationships. These representations need not correspond to familiar genre categories. A system may learn that recordings work well in a certain sequence even when they differ in instrumentation or historical origin. It can then rank candidates according to predicted actions such as completing a track, saving it, returning to the service, or avoiding an immediate skip.\n\nThese signals are ambiguous. Skipping may indicate dislike, an interruption, a search for a particular passage, or a preference unsuitable to the current setting. Long playback may reflect close attention or background use. Shared household accounts can combine several people’s tastes, and entirely new recordings present a “cold-start” problem because they lack behavioral history. Services can address this with audio analysis, supplied metadata, editorial information, and controlled exposure, but none gives direct access to a listener’s intentions.\n\nEditorial selection remains important alongside automation. Human programmers can organize releases around scenes, genres, events, and historical themes that are not reducible to similarity scores. Playlists may be designed for focused listening, exercise, sleep, or social settings. Their sequencing can draw on tempo, instrumentation, mood, or lyrical content, but these categories are interpretive rather than objectively equivalent across listeners and cultures.\n\nRecommendation also creates feedback loops. Prominent placement generates plays, and those plays can become evidence for further placement. Evaluating a system therefore requires more than measuring clicks or listening duration: exposure, catalog diversity, repetition, and discovery of unfamiliar artists may produce different results. A recommendation that increases immediate engagement is not necessarily the one that maximizes historical breadth or introduces the greatest number of new performers. The objectives chosen by the service are part of how its catalog is organized and experienced."
    },
    {
      "heading": "Listening culture, artistic practice, and international circulation",
      "text": "Streaming has changed the unit around which much everyday listening is organized. Albums remain important, but individual tracks can circulate independently through search, playlists, social sharing, and automated continuation. A listener may encounter a song without seeing its original cover, track sequence, or release context. This separates access to a recording from some of the packaging and explanatory material traditionally supplied with physical editions.\n\nThe effects differ across musical traditions. In popular music, tracks often function as readily identifiable items with a principal artist and title. In classical music, listeners may need to distinguish the composition from a particular performance, edition, conductor, ensemble, and soloist. A single work can occupy many tracks, while an album can contain several works. Jazz catalogs present related challenges involving alternate takes, session dates, changing personnel, and reissues. A uniform interface can simplify browsing while obscuring information important to these traditions.\n\nStreaming also extends the usable life of older recordings. A newly encountered catalog track can be played immediately rather than located in a secondhand shop or ordered as an import. Film, television, social video, and public events can redirect attention toward recordings long after their first release. Availability is not guaranteed, however: unlicensed recordings, incomplete transfers, and missing metadata can make culturally significant material difficult to find despite the apparent breadth of a service.\n\nArtists and labels have adapted release practices to an environment of continuous availability and measurable response. Singles, advance tracks, collaborations, deluxe editions, and staggered releases provide repeated opportunities for attention. Some streaming charts count activity from individual tracks toward album-equivalent units, although methodologies vary. Such accounting should not be confused with direct observation of complete-album listening: an equivalent unit is a conversion rule, not a claim that the album was heard in sequence.\n\nClaims that streaming has universally shortened songs or eliminated instrumental introductions require caution. Counting rules and the importance of early retention may create incentives, but song length also depends on genre conventions, radio formats, artistic decisions, and broader historical changes. A statistical association between a platform’s growth and a musical feature does not by itself establish causation. Long-form ambient recordings, extended improvisations, and complete operas also remain available within the same distribution system.\n\nInternational circulation has expanded as recordings become accessible across many territories at once. K-pop, Latin American popular styles, and numerous regional scenes have reached audiences through combinations of streaming, video, touring, and fan activity. Yet global access does not mean equal visibility. Language, local licensing, editorial staffing, marketing resources, payment access, and mobile-data costs continue to shape what listeners encounter and which audiences services can reach."
    },
    {
      "heading": "Measurement, platform governance, and unresolved problems",
      "text": "A stream is an operational measurement, not a universal natural unit of listening. Services and chart organizations establish rules about minimum duration, repeat activity, account eligibility, and excluded behavior. Spotify, for example, generally counts a song stream after at least 30 seconds of listening. Such a threshold helps define an event for reporting; it does not mean that every qualifying play receives an identical payment or represents equal attention.\n\nMusic charts transform these events into rankings using their own methodologies. Paid and advertising-supported listening may receive different weights, while album-equivalent calculations translate track activity into composite units. Rules can change as listening practices change. Comparisons across services, countries, or historical periods therefore require attention to what was counted and how. Monthly listeners, followers, total plays, and subscribers describe different populations and should not be treated as interchangeable measures of an artist’s audience.\n\nArtificial streaming attempts to generate activity that does not represent genuine listening demand. Methods can involve automated accounts, repeated plays, or coordinated manipulation. Because allocation and chart position depend partly on measured activity, fraudulent streams can divert money and distort discovery signals. Services and distributors use account analysis, anomalous-pattern detection, withholding of payments, and removal procedures to address the problem. Distinguishing manipulation from legitimate repetitive listening nevertheless requires more than observing that a recording has many repeats.\n\nSynthetic audio introduces related but separate questions. Artificial intelligence can assist composition and production, generate recordings, or imitate vocal characteristics. A recording’s eligibility for distribution, the rights in its underlying material, consent concerning a recognizable voice, and whether its plays are authentic are distinct issues. Labeling something “AI-generated” does not by itself resolve copyright ownership, contractual permission, or fraudulent use. Legal treatment varies by jurisdiction and continues to develop.\n\nPlatform governance also includes handling ownership conflicts, impersonation, prohibited content, and account sanctions. Large services can influence access through search placement and contractual terms even without owning the recordings. Competition authorities have examined relationships among streaming services, major rightsholders, app stores, and device ecosystems. Questions include the effects of exclusivity, restrictions on directing users to alternative payment methods, and unequal access to promotional tools.\n\nListening records raise data privacy issues because they can reveal patterns of daily activity and personal preferences. Services collect data for playback, personalization, advertising, security, and reporting, subject to applicable law and their policies. Preservation poses another difficulty: a vast current catalog is not equivalent to a permanent archive. Continued access depends on functioning accounts, software, licenses, and accurate records. Energy use likewise extends across networks, data centers, and listening devices; credible environmental comparisons must specify the device, listening duration, network, and alternative format rather than assign a single universal footprint to every stream."
    }
  ],
  "links": [
    "music",
    "internet",
    "intellectual-property",
    "copyright",
    "recommender-system",
    "machine-learning",
    "tempo",
    "popular-music",
    "classical-music",
    "jazz",
    "causation",
    "k-pop",
    "artificial-intelligence",
    "data-privacy"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}