In recent years, the phrase Artificial Intelligence (AI) has become part of our everyday conversations. In particular, since the explosion of generative AI models like ChatGPT or other models that can generate high-quality audio (e.g. AudioGen, AudioLDM 2, and Stable Audio Open, which has just been released and is trained with Freesound sounds), AI systems have demonstrated being capable of doing things which were unimaginable only some years ago. We don’t know how AI will impact society at large, but it is very clear that AI is here to stay and that it will more and more become an important element in our lives. This is raising concerns, for example, within the artistic community, AI can be seen as a potential threat to the way in which many artists make their living, and one that questions the values of effort, skill and creativity. Concerns about the potential impact of AI on a platform like Freesound have also been expressed in the Freesound forums. Here, at the Music Technology Group of Universitat Pompeu Fabra where Freesound happens, we have also been discussing such concerns. For those who don’t know, the Freesound platform was started (and is still developed and maintained) in the context of a research group at a public university, and the open nature of the platform makes it ideal as a resource for audio research, not only for our university but for hundreds of other universities, research centres, companies and individuals all over the world. The relationship between Freesound and AI has therefore been existing since the beginning of Freesound, but only recently, with generative AI, AI has become an object of extensive public debate. For all these reasons, we have decided to write a blog post to state our position in the era of generative AI, to discuss how we think Freesound stands in relation to AI threats and opportunities, and how we expect Freesound to become a good example of AI being used as a tool to benefit the community. So, with no further ado, let’s get started!

The threats of AI

We would most likely all agree that the idea of AI as a threat to a community like Freesound manifests because of the developments in generative AI, which bear the idea that recording and uploading sounds might become redundant because we’ll be able to generate all the sounds that we want by feeding prompts to an AI model. The way we see it, such models pose two specific threats which could be referred to as AI sound flooding, and the replacement of Freesound by AI. Let’s discuss them separately and see which actions could be taken to mitigate potential issues derived from these threats.

AI sound flooding

If AI gives us the capability to quickly generate thousands of unique sounds using generative models, it could happen that huge amounts of AI-generated content get uploaded to Freesound, diluting human-crafted sounds and reducing the overall quality of search results. Even assuming generative sound AI systems reach a quality which is comparable to human-crafted sounds, some of these sounds will be fake. For example, think of an automatically generated “crowd ambience in the streets of Barcelona”, we would not want users to believe these are actual real recordings from Barcelona, even though the sounds could be of good quality and reusable. Also, sounds generated with AI might be accompanied by AI-generated descriptions which could be inaccurate or lack relevant contextual information that only real authors could provide. Finally, AI-generated sounds will most likely present a much higher level of homogeneity than original sounds because these are severely restricted by the capabilities of the generative models that produced them. Altogether, an AI sound flooding would lower the value and quality of Freesound, and that would have an obvious negative impact on the community.

A way to address such potential problems is to make sure that the AI-generated content which gets uploaded to Freesound also gets flagged as such. Having content flagged would allow us to implement filters and highlight, if need be, the AI nature of sounds. Flagging could be achieved by adding an option to mark sounds as being “AI generated” in the sound description form, but also by using systems that will automatically identify content which was generated with AI (e.g. using watermarks or other methods which will be, ironically, AI-powered). In fact, the idea of being able to identify AI-generated content is a core concept of the recently approved EU AI Act, and a very relevant research topic in itself.

That said, let us mention that we don’t see the flooding problem as a very likely scenario, at least in quantities that would severely impact Freesound. We believe that, unlike other platforms in which users upload content and get economic rewards for its consumption, Freesound users have no incentives for uploading large amounts of content and abusing our terms. Furthermore, the relatively small size of Freesound plays in our favour: in case of need, we might be able to impose more strict rules for uploading sounds and manual checks which would allow us to avoid content flooding.

Replacement of Freesound by AI

The other big threat of generative AI in relation to Freesound is the potential of AI to become a replacement for Freesound in the sense that people looking for sounds might not need to visit Freesound if their AI assistant directly creates the sounds they need and these sounds are of good enough quality. This might result in less overall activity on the website (less traffic, fewer downloads, fewer ratings, fewer comments, etc), and possibly fewer sounds being uploaded and fewer user donations, which would affect the richness of the Freesound community.

Nevertheless, we will all agree that generative AI models can only emulate a small bit of the experience of searching and sharing sounds within a user community. Freesound is much more than just a collection of audio files, and all the social activity and personal stories are something that would not be replaced by an AI generative model. It is difficult to estimate what the impact of a really good generative model could have in terms of Freesound website activity, yet it is plausible to believe that a significant chunk of the Freesound users who only download sounds and are not interested in the community aspects of Freesound, might not feel the need to use Freesound again. But it is also true that Freesound would remain a unique sound-sharing website for thousands of users around the world, and its added value when compared to AI-generated solutions would still be enormous.

We believe that to address the threat of AI replacing Freesound, we have to put special emphasis on the community aspects, for example, implementing new features that promote activity within the community. We also have to continue carrying out research activities that allow us to deploy state-of-the-art tools for searching and discovering sounds within Freesound. By maintaining such a competitive advantage over the experience of generating sounds with AI, we don’t think AI will become a real threat to Freesound’s sustainability.

Can my sounds be used to train AI models?

Another common discussion in relation to generative AI models is that of the restrictions that apply to training models with content released under Creative Commons (CC) licences. This is a complicated matter, and so far there are no clear and unambiguous answers to many of the questions that might arise. Nevertheless, we’ll try to answer the general question “Can my sounds be used to train AI models?” for the different licences with which sounds can be published in Freesound:

• Creative Commons 0 (CC0): this licence basically states that the sound is released to the public domain without copyright restrictions. Therefore, sounds under CC0 can have no copyright-related usage restrictions for training and distributing both predictive and generative AI models (note that we use the term predictive here to refer to AI models that do not generate any kind of content).
  
  
• Creative Commons Attribution (CC-BY): the “BY” clause of this licence specifies that, if a particular sound is used to create some other work (an adaptation or derivative work), attribution should be given to the author of that sound. The first question therefore would be: should an AI model be considered an adaptation/derivative work of the sounds used to train it? Even though there is so far no clear legal answer to this question, there is no doubt that an AI model is based on the data used to train it, and the data is what determines the characteristics of the model. Therefore, we recommend that AI models (predictive and generative) only use CC-BY sounds for training if, on distribution or publication, they provide attribution. This attribution should be given at least to all CC-BY sounds included in the training set, and should be accessible through a public URL containing the list of sounds. However, in the case of generative AI models, what happens with the sounds that are generated? So here comes the second question: should sounds generated by an AI model also be considered adaptation/derivative works of the sounds used to train the model? Answering this second question is even harder than answering the first one, and the answer could depend on the nature of the actual sounds being generated. It could be the case that a model generates sounds which only feature small transformations with respect to individual sounds of the training set, or sounds that reproduce entire chunks of training sounds. In that case, the generated sounds could be considered reproductions or adaptations/derivative works of the sounds in the training set. This is what happens when models have memorisation issues, and this is something that AI researchers try to avoid. However, the most common case is that a model generates sounds which don’t have clear individual references in the training set, and in that case these will not be considered adaptations/derivative works. If a generated sound was to be considered an adaptation/derivative work, then, wherever the generated sound was used, attribution should be given to the CC-BY sounds used to train the generative model. But if the sound was not considered an adaptation/derivative work, then no attribution would be required. We therefore think that, as a general rule, sounds generated by AI do not need to provide attribution to the individual sounds of the training set of the model. Nevertheless, we believe that a best-practice for AI models trained with CC-BY sounds would be to recommend their users to provide attribution to the models themselves in the works they create, that is to say, to be transparent about the fact that the a work includes AI-generated sounds, and which models were used to generate them. For example, model creators could recommend their users to use a wording like this: this work uses sounds which were generated using the AI-model X.
  
  
• Creative Commons Attribution-NonCommercial (CC-BY-NC): sounds released under the CC-BY-NC licence provide the same copyright restrictions as sounds with CC-BY, with the addition that those sounds should not be used for commercial purposes. Therefore, what we have written for CC-BY applies to CC-BY-NC as long as the AI models are not used in a commercial setting or for a commercial purpose (as defined by the NC term of the Creative Commons licence). In a commercial setting, CC-BY-NC sounds should not be used for training  AI models (both predictive or generative), and AI models trained with CC-BY-NC can not be distributed.
  
  
• Sampling Plus (CC-Sampling+): even though CC-Sampling+ is a legacy licence and its use is discouraged by Creative Commons since 2012, it was used at the very beginning of Freesound and there are ~11k sounds that bear it. For what AI is concerned, we interpret the CC-Sampling+ licence in a very similar way to CC-BY, as it does require attribution similar to the “BY” clause. Even though the sharing of the sounds as-is in a commercial setting is not permitted in CC-Sampling+, this is not the case when distributing AI models, and therefore no commercial limitations would apply. Therefore, our recommendation is that AI models can use CC-Sampling+ sounds as long as attribution is provided in a similar way to CC-BY sounds.

One important thing to mention however is that all the above restrictions are only concerning copyright. There are no copyright restrictions on non-copyrightable works, and there are exceptions in the copyright law that could allow the use of copyrighted works without permission of the copyright holders. In addition, there might be other considerations beyond copyright (e.g. database rights, privacy, security) that might affect decisions of whether it should be allowed or not to use some specific works for training AI models. Therefore, CC licences are not the only relevant thing to take into account to answer the question “Can my sounds be used to train AI models?”. For those who want to know more, you can read this blog post published by Creative Commons, and follow extra links there.

In summary, if you are a Freesound user and you are concerned about your sounds being used for AI, the most restrictive choice you have is to use CC-BY-NC to avoid having your sounds used for training commercial models. Alternatively, a middle option is to use CC-BY which will allow commercial uses but will, at the very least, help enforce transparency of AI models by promoting a disclosure of the training set (or part of it).

How can AI benefit Freesound?

So far we only talked about AI threats and restrictions for training and distributing AI generative models, but there is indeed a lot of good stuff that AI has to bring, so let’s now look into that. 

First of all, let us remind you again that Freesound is a project that was born in a research group of a public university, and we, at the research group, have been working on AI for many years. Freesound is indeed a very valuable resource for AI research all over the world, an indication of that being the over 850 research papers citing Freesound. This has a tremendous impact on all audio-related research fields. AI models (or machine learning models as we used to call them before the AI boom) trained on data from Freesound are being used for many research tasks such as automatic sound identification and localization, speech recognition and acoustic scene classification (to name a few). In fact, AI models are already powering some Freesound features like sound similarity and tag recommendation, and will power other features with which we have experimented but which have not been deployed. Furthermore, the Freesound community is often invited to participate in research experiments through which feedback is collected that informs research and that eventually allows the development of new features that are deployed in Freesound and benefit the community. The Freesound community and predictive AI research are already a good example of a synergetic ecosystem. 

However, it could be argued that the challenge comes with generative AI. While the threats with generative AI are real and we have to assume that some things will change in the future, generative AI also has a great potential to benefit the community. Using AI tools that generate sounds, we can explore new sonic realms which were not available before, and new forms of art, aesthetics and sound sharing can emerge. Generative AI could also be used for sound transformation instead of from-scratch generation, and this could be a very useful sound design tool for the community which could potentially be integrated into Freesound. With such a tool we might be able to transform sounds in a way that would allow us to make footsteps “heavier”, or an explosion “softer”, or make a “car sound like a dolphin”, or separate mixed events in the same sound. And these are just some examples. 

The overall question is how to use and integrate AI technology in a way that benefits the Freesound community, and this is what we are starting to address with this blog post. At the Music Technology Group, we conduct cutting-edge AI research while considering its social, economic, legal, ethical, cultural, and artistic implications. Research is being carried out, not only by us but also by many other research institutions, which provides new ways to address the threats raised by AI, contributing to a fair and responsible AI. And, again, Freesound is a very important resource for such research. In a similar way in which Freesound became a great example of a sound-sharing ecosystem based on Creative Commons licences, we have the opportunity now to become an inspiring example of how AI can be used to empower a community and to articulate positive experiences for sound practitioners alike.

Let us finish by saying that the contents of this blog post mainly represent the opinion of the Freesound team of the Music Technology Group, but have also been shared and discussed with the team of Freesound moderators who have provided great insights. Also, we are very thankful for the comments from Malcolm Bain at Across Legal, Barcelona.

Thanks for reading until the end, and please let us know what you think in the comments section forum thread that we created for this topic 🙂

…the Freesound Team

Hi everyone!

Another year has passed and here we are with our 2023 Sustainability Report! In this post you’ll find an update about how 2023 went for Freesound in terms of sustainability, and also we’ll let you know about our plans for 2024. Most of the information in this post is very similar to that of previous years, but you might still find it interesting :). The report is split in a number of sections discussing specific aspects that contribute to the sustainability of Freesound, and a final section with a summary, conclusions and future perspectives.

Sound contributions

Sound uploads are an essential part of the sustainability of Freesound. In 2023, 40,940 new sounds were uploaded to Freesound (which corresponds to 1,134 hours of audio). Even though the number of new sounds is slightly lower compared to the previous year, the actual number of hours of audio has increased (70 more hours). Last year we surpassed the 600k sounds mark, and we don’t foresee any problems for Freesound in terms of sound contributions. In addition to the number of uploaded sounds, it is also worth mentioning that 2023 saw a 16% increase in the number of sound ratings (226k), which we believe should be attributed to the deployment of the new UI and the way in which rating widgets are displayed. If you’re curious about these kinds of stats, you can check the 2023 in numbers blog post which was published a couple of months ago.

User donations

Last year we reported a decreasing tendency of user donations during 2022, which we still understood as an effect of the “get back to normality” effect after the COVID19 lockdowns which resulted in almost a 30% increase in donations. In 2023 this tendency seems to have been continuing, but the final numbers are pretty similar to those of 2022. In 2023, we received 44,100€ from 5,200 individual donations. Nevertheless, after the deployment of the new UI in late October, we started to observe an increased number of donations which we expect to consolidate during 2024, and revert in this way the decreasing tendency. The income from donations has been spent on Freesound maintenance and development efforts. In particular, we put a lot of emphasis on finishing the new user interface, and on significant backend changes related to the search engine which will set a new ground to allow us deploying new search-related features.

Contribution from UPF

Freesound is an initiative of the Music Technology Group, a research group of the Universitat Pompeu Fabra (UPF, Barcelona, Spain). In 2023, the UPF contributed to Freesound similarly to previous years. The UPF has increased the capacity of our Kubernetes cluster, which currently consists of 98GB of memory and 35 CPUs for the Freesound website, with an external server used for serving sound downloads and some static files (which transfers around 2TB of data per day). Together with the infrastructure, the UPF also provides some maintenance and IT support (thanks Javi, Miquel and everyone else!). Furthermore, a big part of the human costs for maintaining the Freesound website and related research activities (i.e. our salaries), are a contribution by UPF. In relation to that, during 2023 we have been able to open yet another PhD position that deals with research topics related to Freesound, increasing our machine listening research team. Read the next section for more information about that.

Contribution from research grants and Freesound-related research at UPF

Research activities related to Freesound have been intensified during 2023. We have been able to partially fund such research through different projects and funding agencies including Ajuntament de Barcelona (SoundLights project, BIT Habitat-La Ciutat Proactiva 9382417), Secretaria d’Universitats i Recerca del Departament de Recerca i Universitats de la Generalitat de Catalunya (Joan Oró program 2023FI-100252), Ministerio de Ciencia, Innovación y Universidades (Maria de Maeztu Strategic Research Program CEX2021-001195-M / Musical AI project PID2019-111403GB-I00), Ministerio de Asuntos Económicos y Transformación Digital (UPF-BMAT Chair on Artificial Intelligence and Music, TSI-100929-2023-1). The research topics in which we’ve been working on mainly include:

• Further support of Essentia, the audio analysis library that powers Freesound sound analysis.
• Research on methods for automatically classifying audio events and development of artificial intelligence models.
• Research on the definition of a sound taxonomy that we expect to start adding to Freesound in 2024. We carried out an evaluation in which many Freesound users participated.
• Research on new strategies for sound similarity.

If you’re interested in learning more about the research that happens around Freesound, not only at the MTG but also around the world, be sure to check the papers section of the Freesound Labs website. You’ll see that in 2023 alone, there were 195 research papers referencing Freesound!

Commercial usage of the Freesound API

Freesound has an API endpoint which allows third parties to develop applications that incorporate Freesound content. Usage of this API is free for non-commercial purposes, while commercial use of the API requires a commercial license. In this way we make sure that commercial applications using Freesound also contribute back to the community. Note that this is independent from the license of the sounds themselves, which need to be respected regardless of the API usage agreement. In 2023 we maintained a similar number of license agreements, maintaining the yearly income around ~6,000€. We spent this money in the same development efforts described in the User donations section above.

Summary and perspectives for 2024

In 2023 we have been able to increase research and development efforts with respect to 2022. The research team has been further consolidated and we have prepared Freesound’s technology backend to accommodate new research outputs. We expect user donations to increase during 2024 as a result of the deployment of the new UI, and we also expect to be able to further increase the research team. In terms of development, we expect to focus our biggest effort on i) further fixes and improvements for the new UI, ii) new features for the search page, and iii) a new feature to allow the creation of collaborative sound collections.

We’d like to finish this post by saying thank you to everyone who contributed to Freesound during 2023: sound uploaders, downloaders, moderators, people who engage in forum discussions and in Github, people who donated, and everyone who is part of the community or is related to Freesound in some way. We’ll let you know how things go next year in the 2024’s sustainability report!

frederic, on behalf of the Freesound Team

Hi everyone,

We are happy to announce that we have just released a new feature that will allow you to display the results of a search query on a map. Under the “advanced search” settings panel, you’ll now find a new option to “Display results in map” which will, well, display the search results in a map :). The query can then be further refined and filters applied just like when displaying results in the usual way. Before this feature was introduced, it was only possible to filter the sounds of a map with a single tag or username (although the way to do that was not well documented). This is still possible, but now you’ll be able to do much more using all the tools of the search page. For example, have you ever wondered how ambulances sound in different parts of the world? Or how do kids sound when in the playground?

When “Display results in map” is enabled in the search page, you’ll see a map of sounds and also a link to open the same search results in the usual map page (where the map takes the full width). There you’ll find the option to embed the map, which is now compatible with search-based maps. For example, here is an embedded map with the ambulance sounds:

…aaaand this is it for today, we hope you enjoy this new feature!

frederic
on behalf of the Freesound team

Hi everyone,

It happened again! Another year has passed so it is time to show you some statistics about last year’s Freesound activity. As usual, we start by showing some general statistics similar to those shown in previous years’ posts, and at the end we focus on a different aspect of Freesound, which this year has to do with sound ratings. Without further ado, the number of new sounds uploaded during 2023 has been of…

40,940 new sounds!

which corresponds to…

1134 hours of audio!

There have been ~9,000 less uploaded sounds during 2023 when compared to 2022, however, in terms of hours of audio, there have been 70 more hours of audio uploaded. This is most likely due to more field-recordings being uploaded during 2023 (see below), which has increased the average duration of sounds to 100 seconds per sound.

Here is the Creative Commons license distribution of these newly uploaded sounds:

The percentage of CC0 sounds is the same as in 2022, however, the percentage of Attribution NonCommercial sounds has been increased by 5% (in detriment of Attribution sounds). In 2022, we observed that the distribution had shifted a bit in favour of using CC-BY sounds (and in detriment of CC0), but this year we see how the use of CC0 goes back to what it used to be before 2021 and it represents two thirds of the licenses used for new sounds.

With the new additions from 2023, Freesound now currently hosts an amazing total of 616,881 sounds (we surpassed the 600k sounds mark!), for a total audio length of 429 days and 14 hours. Here is the evolution of the total number of sounds since the beginning of Freesound, and the prediction for the future that we made last year:

Following last years trend, the total number of sounds is growing slightly below the prediction. This is not only due to less sounds having been uploaded in 2023, but also because a major contributor of Freesound decided to remove his account and all his sounds (8k sounds 🙁 ) which all in all results in a significant drop with respect to the expectations. It looks like our earlier predictions of reaching 700k sounds in 2024 will not be possible, but hey, we’re still doing pretty good at capturing the sound of the world!

Here is a tag cloud of the tags of the sounds uploaded during 2023:

As every year, the usual popular tags don’t change much (field-recording, music, loop, synth, …). We expected to see more field-recording related tags that would explain why the average sound duration was increased but if we look at the numbers we see that the percentage of new sounds with tag field-recording, ambience and ambient is very similar to that of the last year. Therefore the increase of the average duration might not be due to any general trend but to the impact of Philip_Goddard’s 237 hours of uploaded content (split in 251 sounds, roughly 1 hour per sound)! Here are some of the best rated long sounds uploaded this year:

And here is the classic chart of the users who have contributed the most sounds in 2023:

Thanks everyone (not only those appearing in the table) for all the contributions! It is absolutely incredible to see that many new and high-quality sounds being uploaded every year 🙂

And what about downloads? The number of sound downloads (including packs) during 2023 was…

21,646,878 downloads!

This is an incredible amount of downloads, even though, again, we recorded slightly lower number compared to the past year. All in all, users have downloaded more than 236M sounds and packs from Freesound!

The term-cloud below shows the most common query terms that have been used when searching in Freesound during 2023:

The top 10 terms are the same as in previous years, with slight variations in the ordering: rain, wind, explosion, music, whoosh, footsteps, woosh, click, scream and piano. If we actually combine the whoosh and woosh, that would bring wooshes to the second position, only 10 daily queries below wind.

Now some extra general statistics: In 2023, 15k messages were sent, 1.5k forum posts, were written, 226k sound ratings were made, 41k sound comments were written. These numbers are following last year’s slightly decreasing trend for the number of messages and sounds comments, but are in fact increasing for the number of forum posts and sound ratings. Regarding the forum posts, the growth could probably be attributed to the intense discussions related to the beta testing and release of the new Freesound user interface which happened earlier this year, in October 2023.

To finalise this post, lets show some numbers about sound ratings. If we consider ALL sound ratings from Freesound and count how many ratings of each value (i.e. 1 star, 2 stars,… 5 stars) there are, we obtain a histogram like this:

As you can see, almost 80% of the ratings are either 4 or 5 stars, meaning that Freesound users generally only rate sounds when they like them. This makes us think that we could eventually replace the rating system by a simple “like” button… maybe something for the future? what do you think? Another thing to look at is, how many ratings do individual sounds receive? Well, approximately 52% of the sounds in Freesound have at least 1 rating, and from those which have ratings, the histogram of number of ratings per sound looks like this:

This is quite expected: ~30% of the sounds with ratings only have 1 rating, ~15% have 2 ratings, and the percentage keeps on going down progressively so only the more “famous” sounds will have many ratings. All in all, in 2023 there have been 226k new sound ratings, which is approximately 15% more than the average number of ratings of the previous 3 years. If we look at the number of ratings per month however, we can see that the overall number of ratings has been increased by ~16% during the last two months, right after the introduction of the new UI. We will see if this tendency gets established or this is just a coincidence, but it looks like the new UI encourages more sound ratings, maybe because it maximizes the number of sounds shown on screen, specially in the front page 🙂

That’s it for this year’s post, thanks for reading and we hope you enjoy a 2024 full of sounds!

frederic, on behalf of the Freesound Team

Hi everyone,

We are very happy to announce that finally, after years of work, we are able to release the new Freesound user interface (UI) codenamed Beast Whoosh (or BW, or Freesound 3). It was early 2017 when we first contacted the UI/UX designer Marc Ruaix to work on a new version of Freesound and prepare it for the future. Our initial intention was to only make a small update of the visual look of the website, but we soon realised that bigger changes would be needed not only on the frontend but also on the backend if we wanted to continue adding new features to Freesound and making its development more sustainable. After the design was finished, we started working on its implementation but only intermittently. We’ve only had a few resources (i.e. hours) to spend on the efforts for the new UI, and the changes that have been finally implemented in both the frontend and backend have been enormous (much more than we anticipated). Fortunately, while implementing the new UI we took the opportunity to also modernise and refactor a lot of the code that runs behind Freesound. And now, after all these years, we’re finally ready to release the new Freesound 🙂

This new UI has been in public beta testing for almost 2 years, and many of you have contributed providing feedback and suggestions which have made it much much better. Even though the essential structure of Freesound remains unchanged with the new UI, you’ll see that it introduces a ton of improvements in terms of workflow and features. Also, it provides new ground for adding more new features in the future. What follows is a list of the most important changes and new features implemented by the new Beast Whoosh user interface:

• Updated overall look and feel to be plain and make navigation easier.
• Responsive design that adapts to mobile phones and tablets.
• The new UI includes a light theme and a dark theme that can be configured in your account settings.
• Different website sections are now available through the upper menus.
• Some detailed information like sound downloaders, sound comments, similar sounds (and more) is now shown using modals which allow you to access that information without having to leave to a different page.
• Added more sounds (and packs!) in the front page.
• The “Random sound of the day” section of the front page has been turned into a sort of game in which the sound name and description is not shown initially so you can guess what the sound is before displaying that information. 
• The new “Manage sounds” page will allow sound uploaders to better keep track of the upload process of their sounds and manage sounds afterwards.
• You can now edit the description of multiple sounds at once, just like when you describe multiple uploaded files at once. You’ll find how to do it in the manage sounds page.
• In the sound description or in sound comments, you can now add timestamped annotations/comments that will render with little play buttons. See an example of this in this sound. You do that by typing some thing like #1:27 your comment to indicate that at minute 1 and 27 seconds, something happens.
• The interface for describing and editing sounds now also includes a sound player so you can listen to the sound while describing/editing its information.
• The new “Charts” page shows some statistics about user activity. This sort of replaces the old “People” page. If you have ideas of other statistics to add to this page, please let us now.
• An option has been added to the account settings to show sound spectrograms by default in sound players (instead of the waveforms). Also, spectrogram and waveforms can be toggled in any player by doing alt+click on them.
• An option has been added to the account settings do enable/disable sound playback polyphony. The default is to enable polyphony, that is to say, two enable multiple sounds playing at the same time (which is how Freesound has worked over the years). Also, even if you have the polyphony activated, you can now do alt+click on the play button on sound players to stop any other sound that was playing and start playing the selected sound.
• An option has been added to the “advanced search” panel to show search results in a grid which allows to get more search results in less screen space (also, in grid mode 30 sounds are returned per page instead of 15). This addresses some concerns raised by users testing the new UI about the number of sounds visible on screen. The preference for grid display is remembered per user account.
• An option has been added to the “advanced search” panel to return only remix sounds in search queries. This will only return sounds that either have been remixed or are a remix of other sounds from Freesound. 
• An option has been added to the licenses filter in the search page to filter results by Free Cultural Works approved licenses (that is to say, sounds under CC-BY or CC0).
• Sound bookmarks are no longer named, but they can still be categorized. Also, bookmarks are no longer public to other users. We’re planning many improvements in relation to bookmarks (including a rename to Collections), so stay tunned.
• You can now bookmark a sound by doing alt+click on the bookmark button, and this will save the bookmark under the last bookmark category that was used without displaying the bookmarking interface. This is a great way of quickly adding bookmarks.
• The “Map” page now has an option to filter by tag (and also embed the maps generated using that filter).
• The “Search” page now has an option under the advanced search options to Display results as packs. This options effecitvely allows to search for packs in the search page.
• The pages to display the full list of sounds or packs by a user now use the search page (with a username filter applied). This allows to further sort, match and filter the list of sounds.
• The “Tags” page now uses the same backend as the search page, which allows to filter by tags but also to apply other filters from the search page.
• The similar sounds option now returns 5 pages of similar sounds instead of only 1.

Note that during some days you’ll still have the chance to switch back to the old UI by using the three dots menu at the top menu of Freesound. However, this option will be disabled in the coming days.

That’s all for now,
We hope you enjoy the new Freesound!

frederic, on behalf of the Freesound team

EDIT: just for the record, I’m adding below screenshots of the Freesound landing page for the original website (back in mid 2000s), the Freesound 2 update (which happened in 2011) and the current Freesound 3 update (2023).

Dear Freesounders,

Welcome to our 2022 Sustainability Report! Here we give an update on how 2022 went for Freesound in terms of sustainability, and present our plans for 2023. You’ll see that most of the information is very similar to previous years’ posts, but still, we think that it is important to share a yearly update with the community. The report is split in a number of sections discussing specific aspects that contribute to the sustainability of Freesound, and a final section with a summary, conclusions and future perspectives.

Sound contributions

Sound uploads are an essential part of the sustainability of Freesound. In 2022, 49,153 new sounds were uploaded to Freesound (which corresponds to 961 hours of audio). This follows a growing trend with respect to previous years (although 2021 saw a spike), and confirms the healthiness of the Freesound community in terms of sound uploads. We are very close to reaching the mark of 600k sounds uploaded to Freesound. But not only sound uploads are important, also sound comments (49k), ratings (174k) and all other ways in which Freesound users interact and generate valuable content for the community. If you’re curious about these kinds of stats, you can check the 2022 in numbers blog post which was published a couple of months ago.

User donations

User donations have continued to fall in 2022 compared to 2021, and are now back again to levels similar to 2019. What we are most likely seeing here is the “get back to normality” after the COVID19 lockdowns which resulted in almost a 30% increase in donations. In 2022, we received 46,600€ from 5,300 individual donations. We expect the number of donations in 2023 to stop the lowering tendency and keep similar numbers as in 2022. As usual, donations income has been spent on maintenance and development efforts. In particular, we put a lot of emphasis on upgrading our backend technology stack (for the nerds: we upgraded from Python 2 to Python 3, and from Django 1.1 to Django 3.2 ), and on the new user interface (which is getting closer and closer). Let us take this opportunity to thank again everyone who donated to Freesound!

Contribution from UPF

Freesound is an initiative of the Music Technology Group, a research group of the Universitat Pompeu Fabra (UPF, Barcelona, Spain). In 2022, the UPF contributed to Freesound similarly to previous years. We have continued our infrastructure improvements which allowed us, among other things, to make Freesound faster and to significantly increase download speeds. The current infrastructure consists of a Kubernetes cluster with 80GB of memory and 25 CPUs for the Freesound website, with an external server used for serving sound downloads and some static files (which transfers around 2TB of data per day). This infrastructure is contributed by the UPF, together with some IT support. Also, a big part of the human costs for maintaining the Freesound website and related research activities (i.e. our salaries), are a contribution by UPF. In relation to that, during 2022 we have been able to open a new PhD position that deals with research topics related to Freesound and which will surely have a positive impact on the platform (more on that below).

Contribution from research grants and Freesound-related research at UPF

Research is at the very core of the Freesound philosophy and, in fact, it is where it all started. We have carried out lots of research activities around Freesound, but, similarly to 2021, in 2022 we did not receive any new big research grant with a primary role for Freesound, but we did receive a 10k USD donation from Google in appreciation for the research efforts on releasing open audio datasets (Google had previously supported our research in that direction and awarded us several Google Research Awards in the past) . Nevertheless, we’ve been able to continue with our Freesound-related research activities which broadly cover these topics:

• Further support of Essentia, the audio analysis library that powers Freesound sound analysis.
• Research on methods for automatically classifying audio events and development of artificial intelligence models. We have started deploying such methods in Freesound and some of their outputs are available through the Freesound API.
• Research on methods to support the sound description process of Freesound, including the definition of a simple taxonomy which will enter an evaluation phase during 2023 (you’ll hear some news about this soon).
• Research on methods for interpretable machine learning in audio classification.
• Research on methods for automatic generation of music instrument samples and music loops.
• Research on hardware and software interfaces for accessing Freesound content (Freesound sampler plugin).

If you’re interested in learning more about the research that happens around Freesound not only at the MTG but also around the world, be sure to check the papers section of the Freesound Labs website. You’ll see that in 2022 alone, there were 161 research papers referencing Freesound!

Commercial usage of the Freesound API

Freesound has an API endpoint which allows third parties to develop applications that incorporate Freesound content. Usage of this API is free for non-commercial purposes, while commercial use of the API requires a commercial license. In this way we make sure that commercial applications using Freesound also contribute back to the community. Note that this is independent from the license of the sounds themselves, which need to be respected regardless of the API usage agreement. In 2022 we maintained a similar number of license agreements (we added a couple but also a couple were cancelled), maintaining the yearly income around ~6,000€. We spent this money in the same development efforts described in the User donations section above.

Summary and perspectives for 2023

In 2022 we have been able to continue focusing on research and development efforts for Freesound in similar way as we did in 2021. We’ll have to keep an eye on user donations during 2023, and think about possible actions if the donations fall below pre-COVID standards. Also, we expect that in 2023 we’ll get new research projects funded which allow us to open new positions and have a significant impact on Freesound. In terms of development, we expect to make the final release of the new user interface during 2023, and continue with the deployment of new technology (particularly classification models). We are also aiming at upgrading the technology we use for the search engine. The new user interface will provide a fresh look to Freesound and some new functionalities, but also represents a big advancement on the technology that we use that will allow us to continue development in a more sustainable manner. Similarly, the backend technology upgrades that we carried out this year are an important step towards Freesound future sustainability.

We’d like to finish this post by saying thank you to everyone who contributed to Freesound during 2022. We’ll let you know how things go next year in the 2023’s sustainability report!

frederic, on behalf of the Freesound Team

[Guest blog post by Riki Refaeli and David Talmor, Freed‘s family and friends]

With great sorrow we part from Friedhelm Hans “Freed” Hartmann upon his tragic death, and provide this tribute to Freed and his many activities including his strong contributions to the Freesound community.

From 1979 to 1986 Freed studied contemporary composition at the Carl Maria von Weber University of Music in Dresden, from 1986 to 1988 at the Akademie der Künste Berlin, and from 1989 to 1993 algorithmic composition at the Institute for Computer Music and Electronic Media at the Folkwang University of the Arts in Essen.  During this time, Freed also joined the German Electroacoustic Society as an active member. From 1993 to 1995, Freed expanded his studies with Itzhak Sadai at the Buchmann Mehta-Musikhochschule of the University of Tel Aviv in the field of music archaeology as a scholarship of the German Academic Exchange Service.  In addition to worldwide performances of Freed’s compositions, which were organized by music companies such as the IGNM or ICMA in various European and non-European countries, Freed’s efforts to introduce new musical ideas to listeners who have less experience with avant-garde or experimental music included the contribution of a large fund of experimental sound textures to the Freesound community, which are still used by several thousand members, the collaborative creation of music pieces, consisting of sounds and sound structures of many Freesound members of different musical backgrounds, musical educational projects and lectures for primary school classes.

Freed’s interest in his music paralleled his interest in the life he built with his life-partner, Riki Refaeli, near the Mediterranean coast just south of Tel Aviv.  Freed and Riki built a life together that included much culture, frequent hiking trips in various regions, and delightful get-togethers with their broad range of friends.  The relationship between Freed and Riki reflected the diversity, color, and activeness of how Freed related to music.  Tragically Freed met his death in the Mediterranean Sea, and although his life ended his works very much continue. Freed’s Freesound profile can be found here. His highest rated sounds in Freesound are found here, and his most downloaded sounds are at here.  You can learn more about Freed’s work at his web site, and on his YouTube channel.

Freed very much enjoyed and benefited from the Freesound community, and we hope that the Freesound community will continue his memory and his approach by continuing to use his works and to remember the very special person from which they came.

– Riki and David

Hi everyone,

I know it has been a while since the last update regarding the new Beast Whoosh user interface that we’ve been working on for the last… 5 years already! Well, now we’re getting reaaaally close to the final release. For that reason, today we’re announcing a pre-release.

When using the old UI, you will see a link in the front page to activate the new pre-release UI for your current Freesound session. You can also activate it following this link. Note that if you open Freesound in a different browser (or in a new private tab), the old UI will appear again. 

Please, provide feedback about the pre-release UI in this forum thread that I just created for the occasion.

The new user interface incorporates many changes and new features. But also, some (mostly minor) things are still remaining to be done. I have added a section in the help page about the new user interface in which you’ll find information about the new features and things not yet implemented (among other things related with the new UI). For conveniency, I also post the list of changes/new features and things remaining here (but be aware that this list here in the forum post will not be updated):

Main new features and changes of the Beast Whoosh UI:

• Updated overall look and feel to be plain and make navigation easier.
• Responsive design that adapts to mobile phones and tablets.
• The new UI includes a light theme and a dark theme that can be configured in your account settings.
• Different website sections are now available thorugh the upper menus.
• Added more sounds (and packs!) in the front page.
• The “Random sound of the day” section of the front page has been turned into a sort of game in which the sound name and description is not shown initially so you can guess what the sound is before displaying that information. 
• The new “Manage sounds” page will allow sound uploaders to better keep track of the upload process of their sounds and manage sounds afterwards.
• You can now edit the description of multiple sounds at once, just like when you describe multiple uploaded files at once. You’ll find how to do it in the manage sounds page.
• The interface for describing and editing sounds now also includes a sound player so you can listen to the sound while describing/editing its information.
• The new “Charts” page shows some statistics about user activity. This sort of replaces the old “People” page. If you have ideas of other statistics to add to this page, please let us now.
• An option has been added to the account settings to show sound spectrograms by default in sound players (instead of the waveforms). Also, spectrogram and waveforms can be toggled in any player by doing alt+click on them.
• An option has been added to the account settings do enable/disable sound playback polyphony. The default is to enable polyphony, that is to say, two enable multiple sounds playing at the same time (which is how Freesound has worked over the years).
• An option has been added to the account settings to show search results in a grid which allows to get more search results in less screen space (also, in grid mode 30 sounds are retruned per page instead of 15). This addresses some concerts raised by users testing the new UI about the number of sounds visible on screen. But also, the spacing has been optimized a bit since the first UI beta (even without using the grid mode).
• Sound bookmarks are no longer named, but they can still be categorized. Also, bookmarks are no longer public to other users. We’re planning many improvements in relation to bookmarks (including a rename to Collections), so stay tunned.
• You can now bookmark a sound by doing alt+click on the bookmark button, and this will save the bookmark under the last bookmark category that was used without displaying the bookmarking interface. This is a great way of quickly adding bookmarks.
• The “Map” page now has an option to filter by tag (and also embed the maps generated using that filter).
• The “Search” page now has an option under the advanced search options to Display results as packs. This options effecitvely allows to search for packs in the search page.
• The pages to display the full list of sounds or packs by a user now use the search page (with a username filter applied). This allows to further sort, match and filter the list of sounds.

Things still missing in the new UI:

• Moderation pages (this is the biggest part remaining, sorry moderators!)
• Modals for listing sounds/packs downloaded by a user, and users who downloaded a sound/pack.
• Remix groups pages (the pages that show sounds that have been remixed)
• Make new sound embeds using the new design
• API credentials management page (for developers only)
• Implement the ruler in the big sound player
• Improvements in accessibility (compatibility with screen readers). Here we’ll need help from the community to find the parts which have no good support for the screen readers.

That’s it for now! I hope you enjoy using the new UI 🙂 We expect to continue working on it during the coming weeks, and are aiming at a final release during Summer 2023. At that time, the option to use the old UI will be removed.

Thanks for reading, and remember to leave your feedback about the new UI in the forum thread linked above.

frederic
on behalf of the Freesound team

[Guest blog post by Craig Smith]

The “SSE” sound effects come from the Sunset Editorial collection which was donated to USC Cinema in 1990. Sunset Editorial had a low-key history in Hollywood. Not a lot is known about them. This is because their credits in films were usually just “Sound Effects — Sunset Editorial”. The company was active from about 1964 to 1987. They mainly did episodic television shows like “Bewitched”, “I Dream of Jeanie”, “The Partridge Family”, and “The Waltons.”  They have 569 credits on IMDB! A lot of the 1960s editing was done by Fred Brown. They were owned in the ’70s/’80s by sound editor Gene Corso.    

Sunset Editorial was a very simple operation. They only edited sound — no mixing. And they worked on a tight budget. Sound editors came and went. There wasn’t much money available for creating new effects, so they depended on effects that sound editors had been using for decades. Freelance editors brought their own collections, and then copied effects to take as they went on to their next job.

Their sound effects collection consisted of dozens of cardboard boxes filled with small rolls of 35mm magnetic film. Each had a slip of paper around it describing the sound.

The rolls were clear 35mm acetate film with two magnetic stripes. One stripe contained the sound, and the other was just to keep the film’s thickness even from edge to edge. Otherwise it would wind unevenly onto a reel.

 At USC Cinema, the sounds were transferred from 35mm mag to 1/4” full track tape. The work was done mostly by Sound Department T.A.s Cormac Funge and Tim Maloney in 1990. I actually tracked them down and was pleased to learn that they’re both still film sound designers!  

Cormac & Tim’s memories of this project were not exactly happy ones. There were about 1,200 rolls of film. With all the handling, it took them about six months to transfer a total of 21 hours of sounds. But they did a great job. 

I got the SSE tapes from the USC Archive in 2016. It was immediately clear that these tapes had a big problem. They were recorded onto used Ampex tape from the 1980s. Tape manufacturers changed their formulations in the early ’80s, and it turned out these new tapes were very unstable. They started to display what became known as Sticky Shed Syndrome.” (Google it.) When this happens, the glue that binds the magnetic oxide to the plastic base becomes sticky, and separates. This makes the tapes virtually unplayable.  

Fortunately, there’s a temporary fix. Tapes can be baked for several hours at a low temperature in an oven. So that’s what I did. Each tape was baked at 150ºF for four  hours, then cooled for four hours.

This made the tapes stable enough to transfer using my Nagra 4.2 full track recorder. 

The good news is that this is an incredibly diverse and rich collection. The bad news is that a lot of these analog 35mm mag elements were copies of copies of copies. So they had a fair amount of noise and distortion. Because of this, I did much more restoration on these sounds than usual. I used iZotope’s RX 10 software. Most sounds cleaned up nicely, but I did eliminate about 20% of them. 

Here’s a very annoying but typical example of the restoration:  

Before:

After:

The resulting 1,022 sound effects are quite good. There are a lot of amazing sounds here ranging from the 1930s into the early ’80s. Check out the ambiances. Even if a sound seems a little odd to you, try layering it with other sounds to create something unique. They can make great sweeteners! To get you started, here’s an incredibly rare recording of the entire Wilhelm Scream recording session:

I couldn’t have done this my self. I want to thank Dino Everett, Ben Burtt, Andrew Kim, everyone at Soundly, Frederic Font, and Lynn Becker.

– Craig Smith

***

Craig Smith has been recording and manipulating sound since 1964. After graduating from USC’s School of Cinematic Arts, he worked as a sound editor and production mixer in Hollywood, specializing in noisy action-adventure films that are blamed for the downfall of society. He left that world in 1986 to teach sound in the School of Film/Video at California Institute of the Arts, where he is now Academic Sound Coordinator.

Craig’s own work experiments with implied narrative and accidental sound design, putting together sounds & images that have nothing to do with each other to create unexpected stories.

Craig is a member of the Society of Motion Picture and Television Engineers, and the Audio Engineering Society.

You can check other posts in the Freesound blog by Craig Smith here: https://blog.freesound.org/?author=14

This was a Freesound guest blog post. Do you have any project or something you’d like to share in the Freesound Blog? Let us know using our contact form