We regularly tell you about progress in artificial intelligence on our blog. For example, in writing, in translation or in voice identification. Today, we are going to introduce you to SoundFilter. It is an artificial intelligence from Google that is able to extract any sound from an audio file.
First of all, let’s briefly recall what an AI is.
In fact, according to futura-sciences.com, artificial intelligence – or AI – consists in “implementing a certain number of techniques aiming at allowing machines to imitate a form of real intelligence” … Which explains why it applies to a number of different topics more and more consequent!
Today, all major IT companies in the web environment are focusing on the challenges, issues and opportunities that are being created around artificial intelligence. Google, Apple, IBM, Microsoft… All of them have their servers and artificial neural networks processing heavy databases.
Therefore, how does artificial intelligence apply in the audiovisual environment and more particularly on audio files?
In this theme, it is also necessary to understand and take into account the specificities of audio. According to Wikipedia, the word “audio” refers to “everything related to hearing and everything that concerns the field of audible electro-acoustic sounds: audio sources, courses, audio books”.
Audio is more broadly the content and media that we listen to, an auditory sensation that is generated by an acoustic wave. Artificial intelligence will therefore interfere and play a role in the processing of these sounds, these audio files. And this for different uses.
Faced with the progress of artificial intelligence, Google is developing tools based on machine learning. In short, machine learning is a technology derived from artificial intelligence that allows computers to learn without having been clearly programmed to do so.
Thus, Siècledigital.fr, tells us that Google has recently developed SoundFilter. SoundFilter is a system based on the learning machine and is able to extract any sound from an audio. Even for a recording, a scrambled audio file, of poor quality, noisy, audibly disturbed.
In addition, the SoundFilter device needs only a very short sound sample to achieve its goal!
The research and operation process of SoundFilter is extracted from a scientific article entitled “Conditional audio filtering of arbitrary sounds at once”. In fact, if the SoundFilter system only needs a short audio sample, it is because it mixes, blends, compares different sources.
In concrete terms, it receives the audio mix to be filtered with an example of the type of sound it must extract at the end. And it is this mixture that requires only one step.
In fact, SoundFilter is trained thanks to the audio samples provided by Google. It then uses a whole artificial neural network and processes wave by wave.
The specialist media company VentureBeat explains that the operation is based on a complex computation by artificial intelligence. The corresponding sound integration is calculated by an encoder.
A conditioned generator takes the audio mixture and the conditioned integration as input. It then produces the so-called “filtered” output.
“The system assumes that the original audio collection is composed of many clips of a few seconds duration that contain the same type of sound throughout the entire duration “… Which is already a considerable work in terms of database and Big Data constitution!
Here are the current results of the researchers around the possible extractions with SoundFilter :
These results are interesting because they already show that a human voice is different from the sound of an object, sound effects or music. Moreover, human voices are infinitely different from each other.
The mixtures are also infinite. This is why SoundFilter can only learn and process sounds and audio sources by experience.
We discover several interesting aspects concerning the use of artificial intelligence for systems such as SoundFilter.
And researchers are far from wanting to stop there in its development! Indeed, by being able to extract – and therefore also eliminate – all types of sounds, SoundFilter could be applied to other sectors and be the pillar of other new technologies.
For example, the researchers explain that their work could be used to create classifiers of audio, sound events.
At Authôt we attach great importance to the development of sound technologies. This is very important because the quality of the original audio file largely influences the final rendering.
Therefore, by improving audio quality again and again, our efficiency is enhanced!
Concerning SoundFilter this is the case. This system does not only extract. What is interesting is that it filters above all! And this, for any sound!
We hope you enjoyed this article about SoundFilter, Google’s intelligent system that is able to extract any sound at once from an audio source. The news and developments around artificial intelligence, new technologies in audio-visual are constantly growing and we are keeping a close watch on them!