Suno AI vs. Lyria 3: AI Music Models Vie for Arabic Song Supremacy

Alanbatnews -

The generative audio AI field has seen rapid growth, with Suno AI and Google DeepMind's Lyria models emerging as key players.

Suno AI focuses on complete structural simulation, while Lyria emphasizes precise waveform modeling. Each has strengths and weaknesses, especially when it comes to Arabic music.

Suno AI's architecture resembles large language models, predicting audio tokens based on training data. It excels at understanding song structure, differentiating between verses, choruses, and bridges, enabling the generation of lengthy, coherent musical pieces.

However, Suno AI can sometimes suffer from digital noise, where instrument frequencies blend with human vocal frequencies, complicating the separation of individual elements later on.

Lyria, developed by Google DeepMind, employs digital signal processing techniques and deep neural networks to directly manipulate waveforms with professional-grade quality.

Lyria excels in sample rate and bit depth, producing crystal-clear audio comparable to studio recordings (44.1 kHz or higher). It also features SynthID, an imperceptible watermark embedded in the audio to protect rights and identify the source, making it a safer option for professionals.

When processing Arabic music, challenges arise in handling quarter tones and complex rhythms. Suno AI benefits from a large database of contemporary Arabic songs, allowing it to effectively mimic the spirit, performance, and vocal expression, particularly in popular genres.

Lyria focuses on the 'physics of the instrument.' When prompted to create an 'oud' or 'qanun' piece, Lyria accurately simulates string resonance, making it ideal for instrumental pieces and soundtracks requiring exceptional clarity.

These applications attempt to emulate Eastern scales by balancing digital frequencies to suit the listener’s taste. While they succeed in delivering coherent melodies, their execution relies on statistical patterns. They excel in simulation but still fall short of capturing the essence of improvisation.

To maximize these technologies, consider these approaches:

For optimal Suno AI results, focus on smart prompts. Use Custom Mode for full control, defining sections with tags like [Intro], [Verse 1], [Chorus], [Bridge], and [Outro].

Describe the music style in detail, specifying the desired emotion, instruments (e.g., acoustic Arabic pop, oud, emotional, slow rhythm, female vocals, high quality). Avoid mentioning famous artists; instead, describe the vocal texture (rough, soft, ethereal, powerful).

If you like the beginning of a song but it ends too quickly, use the Extend feature to add new sections with the same melody and spirit to complete the song. Then, merge all the sections into one long file.

To produce a song in Lyria, focus on the sonic texture. Start with a 30-second intro of improvisations (taqasim), then use the Add Section feature to gradually introduce the lyrics, ensuring the quality of the instruments is maintained without interference.

The key difference lies in Lyria's tonal accuracy and Suno's song length. The world is approaching an era where Lyria will offer APIs to generate complete songs, while Suno improves its audio compression quality. For the Arabic user, the choice depends on the goal: Suno is for outreach and emotion, while Lyria is for professionalism and sonic creativity.