Cracking the International Standard Recording Code

Fabio Vieira[1]

2024-10-16

Abstract

The International Standard Recording Code (ISRC) is a key that uniquely identifies music and music video recordings. In theory, all records released should be assigned to an ISRC. Though this code seems the ideal candidate for random sample generation of music tracks, its assignment workflow is still obscure. The decentralized nature of the ISRC and the lack of public databases make it difficult to obtain access to high quality data. In this text, we discuss the ISRC workflow and its potential limitations regarding its use for generating random samples of tracks. We present our experience in trying to decipher this code and the stumbling blocks we have encountered in obtaining these data from public APIs.

1) Introduction

The music industry can vastly benefit from the calculation of statistical indicators which can provide a better understanding of the business. One example of such indicators is Spotify’s popularity index (Nijkamp (2018), Middlebrook and Sheik (2019)). Data on this index could be used to estimate the average popularity of the typical song in a given population of tracks. Another example is the modeling of the cash flows of tracks which can lead to the development of valuation models for the estimation of a catalogue’s monetary value.

Statistical metrics are usually calculated with the use of random samples that contain characteristics that represent the entire population. Therefore, the generation of statistical samples of individual music recordings can aid in providing powerful insights into the music industry business. However, obtaining these samples is not a trivial task. First, given that little is known about the population of tracks, it is difficult to make any statistical assumptions about it. Second, the lack of publicly available databases represents a big challenge for the gathering of reliable data to compute statistics and perform experiments.

The International Standard Recording Code (ISRC) (ISRC Bulletin 2009-03 (2009), ISO 3901 (2019)), which is a code that uniquely identifies music and music video recordings (Elton 2008), seems to be the natural choice for random sample generation of music tracks. However, there are several issues that remain obscure, particularly regarding what is done in practice when ISRCs are assigned. Moreover, the fact that the ISRC system is decentralized and that there is no central public databases adds up to the difficulties in cracking the intricacies of these codes. Thus, in this text we will discuss these issues and list the problems that we have encountered in dealing with ISRCs.

2) The International Standard Recording Code: the theory

The International Standard Recording Code is a 12 digit code that uniquely identifies music and music video recordings (ISRC Registration Authority (2021), ISO 3901 (2019)). Rights holders assign ISRCs to their recordings for a number of reasons. For instance, some online services, such as Spotify, require an ISRC to be included when a track is uploaded to their system. In addition, Collective Management Organizations (CMOs) require the registration of a repertoire as part of the process of allocating revenues from radio broadcastings and live performances (ISRC Registration Authority 2021). Thus, CMOs can also use ISRCs to help them manage their portfolio of repertoires.

An ISRC has the following structure

AA-6Q7-20-00047.

 Where AA represent the country code, for instance, SK for Slovakia and IT for Italy. The following three characters (6Q7) represent the registrant. These five first characters together make the prefix of the code. The following two digits represent the year of reference, in this case 20 stands for 2020. According to the ISO 3901 (2019), the year of reference is simply the year in which the ISRC was assigned to a record and not the year in which the record was first released or produced. The final five digits represent the designation code. This code ranges from 00000 to 99999 and it is assigned by the registrant. According to the ISRC Registration Authority (2021), every registrant is assigned a range of codes, for instance from 00000 to 00100, and they should assign ISRCs in a sequential order (i.e. 00001, 00002, 00003, …, 00100).

The basic principles of assigning ISRCs are (i) each distinct recording should have only one ISRC and (ii) the same ISRC should not be assigned to more than one recording. This guarantees that the ISRC can be used to uniquely identify a recording and thereby can play a key role in the management of repertoires.

3) Why is it difficult to generate a random sample of ISRCs?

For research purposes, the ISRC can be considered the main candidate for random sample generation. However, given that each ISRC national authority is responsible for managing the ISRCs in a particular country or territory, the ISRC system is decentralized. To our knowledge, there is no public database from where these codes can be obtained. Hence, utilizing the ISRC to generate random samples of music tracks is considerably more challenging than one would assume.

If we had a list containing all the ISRCs available for a particular country, we could simply randomly sample a few of these codes and that would be our random sample. However, given that the ISRC system is decentralized and that there is currently no public database of these codes an alternative must be found. Let us look once more at the code we saw on Section 2

AA-6Q7-20-00047.

 If we were to simply generate a random code following this structure and check on a web API (such as Spotify or MusicBrainz) to see whether this code exists or not, that would not be very efficient, given that the number of possible codes following this structure is enormous. For instance, assuming that only the country code is known, there is a number of the order of  random codes following the ISRC structure, compared to about only about  existing codes[2] for all countries combined. Thus, the sampling space is very large and sparse, which makes it practically impossible to obtain random samples via randomly generated codes.

But, given that the first 5 characters should be known, this number can be drastically reduced. For instance, SLOVGRAM – the joint collective of perfomers and audio and video producers in Slovakia – has sent us a list of all prefixes in Slovakia. This list contains 2168 unique prefixes. Knowing this information can already be extremely helpful to recude the number possible codes. However, we have no idea when those prefixes started to be used or whether they are still active. Thus, that would still require quite some work to figure out this information, since web APIs usually block users from abusing their application by setting a rate limit[3], we would not get very far by attempting to do millions of queries on a public API. Therefore, there are still many limitations in what concerns using ISRCs to generate random samples of tracks.

4) The workflow of assigning an ISRC: the reality

In this section we discuss how an ISRC is assigned according to the ISO 3901 (2019) and ISRC Registration Authority (2021). In addition, we present some issues we have encountered with the ISRCs in practice. A reader should have in mind that our goal is to create statistical metrics for the music industry and that we desire to have these metrics stratified by country.

4.1) Country Code

As described before, the country code is a two capital letter code that represent a country, according to the ISO 3166-2 (2020). An ISRC can be assigned by a rights owner, by an exclusive licensee of the recording or by an ISRC manager. Small producers and small record labels might delegate the workflow of assigning ISRCs to their repertoire to an ISRC manager (ISRC Registration Authority 2021). These ISRC managers might receive the rights to assign codes to only a specific portion of the repertoire or the entire repertoire. The main issue here is that there is no need for an ISRC manager to be in the same country as the rights owner.

Therefore, the country code does not necessarily have a connection with where the music was recorded or produced. Besides, the ISRC Bulletin 2009-03 (2009), that describes how ISRC managers should assign codes, states that national ISRC agencies can apply the guidelines at their own discretion, which makes it hard to know exactly how these ISRC managers work in every country.

In conclusion, the ISRC seems to have the perfect information to allow us to produce random samples stratified by country. However, in reality, this is much more complicated. It is hard to determine where rights owners register their music and it is also difficult to determine to which country a recording should belong once every step in the process of creating, recording, producing, registering and releasing can be done in a different country.

4.2) Registrant Code

The following three digits in the prefix constitute the registrant code, which according to ISO 3901 (2019) is assigned to a registrant by the national ISRC agency. In the case of a rights owner choosing to allow an ISRC manager to register their recordings, the ISRC Bulletin 2009-03 (2009) states that the ISRC manager should request a registrant code directly with the national ISRC agency from where the rights owner is based.

The ISO 3901 (2019) declares that these codes must be unique, but it does not specifically say that one rights owner should have only one registrant code. For instance, we utilized Spotify’s API to obtain ISRCs and extracted the following codes belonging to the Bratislava based label Opus: SKA129020813, CZA239801066, SKA880400012. Thus, in this case, Opus’s recordings are registered with at least 3 different registrant codes A12, A23 and A88. It is a challenge to know how many different registrant codes a particular rights owner might have, which complicates the work of generating random samples that truly represent this unknown population.

4.3) Year of Reference

The two subsequent digits represent the year of reference, for instance 15 represents 2015. The ISO 3901 (2019) states that this should be the year in which an ISRC is assigned to a recording and not the year in which the recording was produced. The ISRC Registration Authority (2021) handbook indicates that the ISRC standard was first published by ISO in 1986. Following this logic, given what is published in the current ISO standard for the ISRC, there should not be codes with year of reference prior to 1986, i.e. 86. However, the ISO 3901 (2019) states that the ISRC registration authority can also decide on other mechanisms to assign the year of reference as long as that ensures that new ISRCs will always be unique.

Figure 1: Distribution of year of reference for sample extracted using Slovak popular words.

Figure 1 displays the distribution of the year of reference in a sample extracted from Spotify’s API using Slovak words. As we can see, there are quite a few tracks with year of reference prior to 1986. We contacted SLOVGRAM who stated that it is common for the rights owner to indicate the year in which the recording was produced, instead of the year in which the ISRC was assigned. Since the ISO give the registration authorities freedom to decide their own mechanisms of assigning year of reference codes, it is hard to say precisely how this is done in practice.

4.4) Designation Code

Finally, the last five digits of the ISRC represent the designation code, which ranges from 00000 to 99999. This part of the code is used by ISRC managers and rights owners to manage their own catalogue. According to the ISO 3901 (2019), every year registrants are assigned a fixed range that they can use to register their recordings. The ISRC handbook states that large registrants might have the entire range at their disposal, whereas small registrants might have access to a smaller range, say from 00001 to 00100 (ISRC Registration Authority 2021). The ISRC Registration authority strongly recommends that registrants assign these codes sequentially (for instance, 00001, 00002, 00003, and so on). However, since it is only “strongly recommended”, there does not seem to be any formal rule on how to register the ISRCs in the assigned range. The ISO 3901 (2019) states that the ISRC registration authority in each country should, for documentation purposes, keep the registrant codes assigned to each rights owner together with the ranges of designation codes assigned each year. But, the fact that there are no public databases with this information and that the existing databases are decentralized, given that every country has their own system, it is hard to retrieve this information. If we wished to negotiate agreements (assuming that is possible) to obtain access to this data, we would have to do it many times over, at least until there is one agreement with every country we are interested in generating statistical metrics.

5) Additional Issues

Elton (2008) has criticized the ISRC system stating that after a code is assigned there is no requirement to submit metadata to the ISRC agency. This has forced some local recording industry associations, which may also serve as local ISRC agencies, to implement processes with their members to centrally compile information about local sound recordings and music video. Thus, this lack of metadata submission requirement reinforces the decentralized nature of the ISRC, making it difficult to obtain those codes and even know how many and where these codes have been registered.

Additional problems include:

• The U.S. and Brazil ran out of their number range and started new country codes.
• International distributors, most notably TuneCore (T.C.), are allowed to register outside of the national registrar system, i.e., TuneCore became a quasi-national registrar.
• The two-letter country code identifies where the registrant’s home office is located. With the industry’s global consolidation, the three majors are now truly unique, representing thousands of sub-labels all over the world; when a U.K. label records (tracks) in France and masters (mixes) in Italy for the global label, the national affiliation is not clear.
• Because of the changes in the industry structure, most European records are not released by recording companies (who own rights) but by the companies of the authors or performers, i.e., the musicians themselves. The number of registrants who need a proper metadata apparatus is very high. (Self-releasing artists.)
• Because self-releasing artists in small countries, like Slovakia, often distribute their music via global specialists based in Austria, UK, Netherlands, Sweden, these specialists play a crucial role in the industry. They take care of the metadata, ensuring the repertoire is often registered in a different country.
• When an artist changes distributors, their music is usually already registered. Thus, some distributors, such as Ditto[4], allow the ISRCs to be manually inputted when the tracks are submitted. This opens the back door for the arbitrary choices of ISRCs that might not follow the ISO standard.

 All these points have direct implications for our work. As we stated in the beginning of Section 4, our goal is to generate statistical metrics stratified by country. The issues raised above show how difficult it is to get a representative sample of tracks from a particular country, given that it is hard to determine where the tracks are registered and the simple inclusion of ISRCs with specific country codes does not solve this problem.

6) Conclusion

Generating statistical metrics for the music industry can be of great value for industry stakeholders, particularly small labels that might not have access to the data-driven solutions that may be available to the big players. These metrics can help acquire a better perspective on the business and provide a wide variety of insights ranging from popularity of tracks to royalty revenues.

Many statistical techniques are built upon the idea of having a sample that represents the population we seek to analyze. These samples are usually obtained in a random fashion, where every element of the population has the same probability of being included in this sample. In the case of music recordings, this is extremely complicated because the population is unknown and retrieving the information that would make this random sample generation possible is impractical.

The International Standard Recording Code (ISRC) seems to provide a solution for this problem, given that every code uniquely identifies a music recording. However, since the population is unknown and that there is no centralized public database of ISRCs, the issue of generating random samples of music recordings remains unsolved. In our experience, given its market potential, rights owners themselves are reluctant to disclose these data.

Therefore, we believe that the best way to improve the transparency of the ISRC system is to create the requirement to provide metadata to the ISRC agency and make a public database available. This could help solving issues regarding royalty payments, since the lack of metadata can make it difficult to determine who the rights owner is, and improve artists overall standings on streaming recommendation systems, given that recommendation system algorithms utilize metadata to generate recommendations (Antal 2021).

References

Antal, Daniel. 2021. “Feasibility Study on Promoting Slovak Music in Slovakia and Abroad.” Open Music Observatory.

Elton, Serona. 2008. “Identifiers Used in the Music Industry.” MEIEA Journal 8 (1): 49–74.

ISO 3166-2. 2020. The International Standard for country codes and codes for their subdivisions. Standard. International Organization for Standardization; British Standards Institution.

ISO 3901. 2019. Information and documentation — International Standard Recording Code (ISRC). International Organization for Standardization; British Standards Institution.

ISRC Bulletin 2009-03. 2009. International ISRC Agency Bulletin 2009/03 – UPDATE: Approval of ISRC Managers to Assign ISRCs . Standard. International Federation of the Phonographic Industry.

ISRC Registration Authority, International. 2021. International Standard Recording Code (ISRC) Handbook. 4th Ed. IFPI Secretariat.

Middlebrook, Kai, and Kian Sheik. 2019. “Song Hit Prediction: Predicting Billboard Hits Using Spotify Data.” arXiv Preprint arXiv:1908.08609.

Nijkamp, Rutger. 2018. “Prediction of Product Success: Explaining Song Popularity by Audio Features from Spotify Data.” {B.S.} thesis.

[1] Institute for Information Law, University of Amsterdam (email: f.v.vieira@uva.nl)

[2] https://www.billboard.com/pro/how-much-music-added-spotify-streaming-services-daily/

[3] https://developer.spotify.com/documentation/web-api/concepts/rate-limits

[4] https://dittomusic.com/en