Feature Request: AI-Powered Harmonic and Arrangement Suggestions

Dear Scaler Development Team,

First, I would like to thank you for creating such an amazing and versatile tool. Scaler has significantly transformed the way musicians approach harmony, chord progressions, and composition. However, I believe there is a potential to take Scaler even further by integrating AI capabilities for advanced harmonic suggestions and real-time arrangement features.

Proposed AI-Powered Features

  1. Dynamic Harmonic Suggestions

Imagine if Scaler could not only suggest chord progressions but also offer dynamic harmonic possibilities inspired by iconic composers, arrangers, or styles. For example:

• If I select GM7, the AI could suggest alternative voicings or complementary chords in a specific style (e.g., “Onno Tunç’s rich string arrangements,” jazz harmonies by Bill Evans, or the modal movements of Debussy).

• Beyond simple diatonic progressions, Scaler could explore modal interchange, chromatic medians, or counterpoint-style harmonic structures.

  1. Arrangement Assistance for Full Instrumentation

A significant enhancement would be for Scaler to offer arrangement suggestions for multiple instrument layers, such as:

• Strings: A counterpoint-based orchestration for violins, violas, cellos, and basses. For instance, if I have a chord like GM7, Scaler could provide suggestions for how to distribute the chord notes across a string section in a dynamic, melodic way (e.g., moving lines, contrary motion, or lush voicings).

• Basslines: AI-generated basslines that complement the progression while considering groove, syncopation, and style. For instance, a disco-style bassline for a progression like GM7 > Em7 > Am7 > D7, or a jazz walking bass pattern for more complex harmonic progressions.

• Melody Suggestions: The AI could suggest motifs or melodic ideas based on the chosen progression and style. This could be particularly useful for genres like jazz, classical, or cinematic music.

  1. Style-Based Inspiration

Users could select a style or artist to inspire the suggestions. For example:

• Disco/Funk: Chord voicings, basslines, and syncopated rhythms inspired by Nile Rodgers or Daft Punk.

• Cinematic/Orchestral: String arrangements and harmonic layers inspired by composers like Hans Zimmer, Ennio Morricone, or Onno Tunç.

• Jazz/Modal: Advanced chord extensions and substitutions inspired by Herbie Hancock or Bill Evans.

  1. Interactive Counterpoint Tool

A dedicated counterpoint feature could allow users to select a chord or progression, and Scaler would generate independent melodic lines that are harmonically and rhythmically interesting. This would help composers quickly generate orchestral or polyphonic textures.

  1. Real-Time Arrangement Outputs (MIDI Layers)

Scaler could generate MIDI layers for various instrumental parts:

• Top-line melodies for lead instruments.

• Inner harmonic movements for strings, keys, or synth pads.

• Rhythmic grooves for drums and bass.

Example Use Case:

Let’s say I am working on a project and I start with a simple progression:

GM7 > Am7 > D7 > GM7

• Strings Layer: Scaler suggests voicings where violins play a high counter-melody (F# > G > A > F#), violas sustain the midrange (B > C > A), and cellos add movement in the low end (G > A > D).

• Bass Layer: Scaler generates a syncopated disco bassline that locks rhythmically with the progression.

• Melody Layer: A melodic motif is proposed for a lead instrument, which evolves over the progression.

Why This Would Be a Game-Changer

By incorporating AI and advanced arrangement capabilities, Scaler could become not just a tool for chords but a comprehensive compositional assistant. It would save time, inspire creativity, and open new possibilities for musicians across all genres.

I truly believe this feature could revolutionize how Scaler is used in the composition process and would love to hear your thoughts on this potential enhancement.

Thank you for considering this idea, and I look forward to hearing your feedback!

Best regards,

Firat Tuncbas (Discorama Music)

4 Likes

Hi @firattuncbas Thanks for the feedback, many of these things considered, some implemented, some on the roadmap. For me it’s important to establish the new ecosystem and get it right. Then one by one introduce some of the features you mention. Would have been nice for release but there is already so much for users to digest.

5 Likes

Si potrebbe attingere a piene mani qui https://github.com/deepseek-ai/DeepSeek-R1
ed anche qui https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero
Lunedì, il laboratorio cinese di intelligenza artificiale DeepSeek ha rilasciato una nuova famiglia di modelli R1 con una licenza aperta del MIT. La versione più grande contiene 671 miliardi di parametri. L’azienda afferma che il modello funziona a un livello paragonabile al modello di ragionamento simulato (SR) o1 di OpenAI in diversi test di matematica e programmazione…

I rilasci hanno immediatamente attirato l’attenzione della comunità AI perché la maggior parte dei modelli open source esistenti è rimasta indietro rispetto ai modelli proprietari come o1 di OpenAI nei cosiddetti test di ragionamento. …

Il modello R1 funziona in modo diverso rispetto ai tipici modelli linguistici di grandi dimensioni… . Cercano di imitare la catena del ragionamento umano mentre il modello funziona per risolvere una query. Questa classe di modelli, che può essere chiamata “ragionamento simulato” o in breve modelli SR , è emersa quando OpenAI ha introdotto la famiglia di modelli o1 nel settembre 2024. …

DeepSeek riferisce che R1 ha sovraperformato o1 di OpenAI su diversi parametri e test, tra cui AIME (un test di ragionamento matematico), MATH-500 (una raccolta di problemi logici) e SWE-bench Verified (uno strumento di valutazione della programmazione).

You could draw heavily here GitHub - deepseek-ai/DeepSeek-R1
and also here deepseek-ai/DeepSeek-R1-Zero · Hugging Face
On Monday, China’s DeepSeek AI lab released a new family of R1 models under an open MIT license. The largest version contains 671 billion parameters. The company says the model performs at a level comparable to OpenAI’s o1 simulated reasoning (SR) model on several math and programming tests…

The releases immediately caught the attention of the AI ​​community because most existing open source models have lagged behind proprietary models like OpenAI’s o1 on so-called reasoning tests. …

The R1 model works differently than typical large language models… . They try to mimic the human reasoning chain as the model works to solve a query. This class of models, which can be called “simulated reasoning” or SR models for short , emerged when OpenAI introduced the o1 family of models in September 2024. …

DeepSeek reports that R1 outperformed OpenAI’s o1 on several metrics and tests, including AIME (a mathematical reasoning test), MATH-500 (a collection of logic problems), and SWE-bench Verified (a programming evaluation tool).

The day that SR (Simulated Rumination) will infest my hobbies, making modern music repetitive and boring just like an increasing amount of text nowadays, I’ll skip all that, and/or I’ll go back listening the good old music made by creative musicians that is in my record library
AMEN

4 Likes