When we found ourselves faced with this example, shared on the website sun, “catchphrases” immediately came to mind as Vittorio Gassman reads the ingredients of biscuits. This time, however, it is not the voice of the “Showman” par excellence to read a text but to do so is aartificial intelligence (AI), which not only applies speech synthesis but is designed to generate music, real unreleased songs.
In the curious example mentioned at the beginning, Suno’s AI receives as input (prompt) part of the text of the MIT license (we talk about it in the differences between open source and free software) and builds a real song on it, which is rather lovely.
What is Suno and how it works
As we have seen previously, Suno is a platform that leverages AI for generate music starting from textual descriptions in natural language. This innovative tool combines the advanced techniques of its generative model with some features that place emphasis on musical creativity. This way users can turn words into notesrhythms and harmonies.
An instrument like Suno bridges the gap between language and music; offers new opportunities for composers, musicians and enthusiasts; makes it easily accessible to all generation of musical pieces captivating and impressive.
The architecture of Suno
Suno uses first of all a text preprocessing module: is responsible for “cleaning” and normalizing the input. It takes care of carrying out the tokenization and eliminating any stop words (which can negatively impact the performance of the piece of music), uses techniques stemming and lemmatization.
Both of these last two techniques are used for normalize the text to reduce complexity and improve accuracy in natural language analysis applications, such as entity recognition, similar word grouping, and text classification. Specifically, it stemming it is the process of reducing words to theirs root form, called “coat of arms”; the lemmatization is similar to stemming but use one morphological knowledge of words and applies linguistic rules to return the correct basic forms.
The preprocessed text is then represented as a sequence of tokens, which serves as input for the next feature extraction step. The following module, in fact, converts the preprocessed text into a numerical representation which can be used for machine learning. It employs different techniques, such as word embeddings (Suno uses Word2Vec or GloVe to represent words like vectors), emotion analysis and topic modeling, to extract the significant features from the text.
This way Suno can capture the traits that describe the semantic relationships, the emotional tone and thematic structure of the user-specified text. From here the pattern generation musical with sequence prediction, harmony and melody generation, instrument selection and orchestration.
At the end of the “chain” there is the rendering audio which deals with transforming musical patterns into an audio format. The final module uses a combination of synthesized sounds and samples to create a rich and expressive audio representation.
Create a song with Suno
To try generating a melody with Suno, simply click on Create on the main project page.
In the box Song description, just specify what you want to achieve: rather than naming artists, it is best to describe in great detail the musical effect you want to achieve. By default, Suno produces a sung song but by activating the option Instrumental an instrumental base can be created.
It is however activating Custom mode which opens up more extensive possibilities than using a powerful tool like Suno.
In the box Lyricsyou can type i lyrics to be translated into music. The service invites you to use exclusively self-created texts, which are therefore not subject to copyright. Otherwise, unpleasant problems could arise. The button Make random lyrics it allows to generate texts pseudorandomly.
In the box Style of music you can set the mood, the mood, the tempo and the vocal characteristics of the singer. Provide a prompt well argued, it means obtaining a well-structured song, which can also be commercially interesting. To get help, you can click Use random style: Suno might be able to surprise you with the perfect genre.
Among the many features offered by Suno, there is the possibility of extend the duration of songs: to merge all the segments into a single song, you can simply click on the “Get the full song“.
The term extension is available on all three plans offered by Suno: Basic (free), Pro e Premier. However, each extension uses 10 credits, such as creating the first song in the chain. The users of the free version they have 50 free credits available every day even if according to the terms of use of the service they cannot use the contents generated for commercial purposes.
Some useful suggestions
Suno was created to manage texts in English but, based on the tests we have carried out, we can say that it does not do badly at all even with Europen. In our language some may be needed optimization more, but the results don’t leave a bad taste in the mouth.
Also check this example: in the box Lyrics not only the text but also detailed indications can be introduced. For example, you can specify sections such as Verses, Chorus, Bridge, to specify the verses, chorus, and bridge section (a link between verses and chorus), respectively. The directions Verses, Chorus, Bridge they can be inserted as they are or enclosed between quadre parenthesis to better highlight them to the AI.
L’Instrumental Break it’s a instrumental interlude which temporarily puts speech aside to give free rein to the orchestra’s pieces. In this regard, as can be seen in the example, specific prescriptions can be provided, so as to obtain the effect you have in mind.
Conclusions
Suno represents a significant step forward in AI-driven music generation, demonstrating the potential of AI for creative song creation and processing.
By transforming textual descriptions into musical compositions, Suno offers a unique and innovative approach to music creation that has the potential to revolutionize the way every user, even those without in-depth knowledge, compose, listen to and interact with music .
As demonstrated by the presentation of tools such as Udio and Stable Audio, there is certainly no shortage of competitors but the latest version of Suno has gained leadership among the most capable solutions in generating high quality musical pieces.