Google DeepMind Releases Lyria 3.5 in Flow Music With Enhanced Vocals and Lyric Generation

Lyria 3.5 Brings Structural Awareness to Flow Music

Google DeepMind released Lyria 3.5 on July 29 in Google Flow Music, marking a generational improvement in music synthesis across four primary dimensions: melodic composition, lyric generation, vocal synthesis, and user control. According to the DeepMind Blog, the update is designed to help creators craft richer, more emotionally cohesive tracks with tighter alignment between generated text and musical structure.

Musicality and Melodic Complexity

The model generates more intricate, natural-sounding melodic structures than its predecessor. This addresses a long-standing limitation in generative music: the tendency toward harmonic simplicity or repetitive phrase patterns. Lyria 3.5’s improved musicality targets users building full-length compositions, where structural variation is essential to listener engagement and commercial viability.

Lyric Quality and Structural Adherence

One of the most significant upgrades is lyric generation fidelity. According to DeepMind, Lyria 3.5 improves both prompt adherence—ensuring generated lyrics match user intent—and structural awareness, meaning the model better understands verse-chorus-bridge scaffolding and syllable counts. This is particularly valuable for music creators working in genres with strict metric requirements, such as hip-hop or pop songwriting, where lyrical rhythm must align with the instrumental grid.

Vocal Expression and Pronunciation

Lyria 3.5 introduces more emotionally expressive vocal synthesis, with improved pronunciation and reduced synthetic artifacts. The model now delivers more varied prosody—intonation, pacing, and dynamic range—making synthetic vocals less recognizable as machine-generated. This enhancement directly competes with tools like Suno and Udio, which have gained traction among amateur producers for vocal realism.

Creative Control Over Tempo and Duration

The release adds granular controls allowing users to specify track tempo and duration before generation, rather than accepting predetermined defaults. This feature addresses a core friction point in generative music: creators often need tracks of specific lengths for video content, podcasts, or streaming-platform requirements. Direct tempo and duration controls reduce post-generation editing overhead.

Why This Matters

Flow Music users now have access to a more compositionally sophisticated model without abandoning the platform’s accessibility for non-professional creators. However, the update is confined to Google’s ecosystem; Lyria 3.5 is not available via API or to competitors building music-generation features. This deepens Google’s moat in consumer music creation, though the closed distribution may limit adoption among professional producers who favor tools like LANDR or BeatConnect that offer third-party integration. If Lyria 3.5’s vocal and lyric quality reaches production-grade standards, it could accelerate music-to-streaming timelines for independent artists, reducing dependency on human recording and vocal coaching—a shift with significant implications for session musicians and voice actors in the music production supply chain.