Template layout:

If the song drifts from your intended sound, reinforce its emotional character in the genre field:

Example: Slow romantic ballad in F major, trumpet & strings, warm 1950s orchestral vibe

These are the most direct ways to cue silence or breath. They work best when placed as their own line or within square brackets between lyrical sections:

✅ Best practice:

If you want the entire track to breathe, and not just vocals, use instrumental tags:

Or:

Pulls back everything for a moment, which works great before a drop or final chorus.

If you want a pause inside a line (especially for spoken-style or ballads), you can break the line like this:

This often results in a slight silence or dramatic breath in the middle of the phrase.

✅ Example (Full Song Snippet):

This is one of the most common challenges in generative vocals (including Songer). By default, models will guess pronunciation from spelling, and if a word isn’t widely known or is spelled similarly to a more common one, or especially when dealing with Latinized names or non-English pronunciations, it will mispronounce it.

You can force the music engine to pronounce it correctly without changing the written lyric by using one of the following compliant phonetic or directive tags.

Add a pronunciation cue immediately before the lyric line that contains the name.

Example 1 - Kolbe

Example 2 - lacrosse

This gently instructs the AI to apply that pronunciation only for that section.

If you want to keep it in line with the lyric to ensure the AI associates it directly with the word:

Example 1

Example 2

Either will guide the model’s pronunciation without breaking compliance; the bracketed term is treated as a phonetic assist, not a lyric.

If the pronunciation tag doesn’t take effect, you can phonetically write the word just for the singing model:

Example 1

AI will read it as Kolbe when sung, but say Kohl-bay in sound.

Example 2

or

It will sing it correctly as “luh-KROSS” but still treat it like “lacrosse.”

Sometimes specifying pronunciation regionally helps:

Example 1

Because Kolbe is German but often used with ecclesiastical (Latin/Italian) pronunciation in hymns, this makes the model read “e” as a long “ay” instead of being silent.

Example 2

That ensures “la” becomes luh, not lah or lah-kraws.

The most effective and compliant method is to tell the model which language to use before a lyric segment.

For example: Bilingual (English/French)

If you mix the languages in one line, specify pronunciation only for the French words:

or

💡 This ensures French vowels stay French even within English phrasing.

Never rewrite “mon amour” as “mohn amoor” — it’ll sound unnatural. Always use real French words with a [French Pronunciation] or [Pronounced …] cue instead.

Example:

If you want to subtly blend accents (like a bilingual singer might naturally do):

or if the opposite (English accent on French lines):

This is fully compliant and can create a more international pop feel.

Examples:

The model now keeps English crisp, French fluent, and the phrasing natural.

Examples:

Yes, you can safely put your language directly in the "Type up to 5 genres or vibes" field. Example:

That’s concise and compliant. Our music engine will read “Australian” as both a regional style and accent hint. It subtly prioritizes Aussie vowel shapes and phrasing.

Instead of “language + accent” (e.g., “Aussie accent”), which may be ignored now, use a contextual and tonal description that defines both the region and character delivery. Use one of these safe, effective variants:

These tags still work if you pair them with authentic Australian context (place names, slang, or cultural imagery) in your lyrics field.

If you’re still getting “Americanized” singing, expand it naturally with contextual cues that fit in 100 characters.

Examples:

Our music engine now pays more attention to context than to accent tags alone.

Embedding Australian identity markers in your lyrics helps anchor the accent model. Examples:

Example Lyrics:

✅ Result: The model now recognizes both context and accent, avoiding UK/US pronunciation patterns.

If the model still slips into a British or US tone, gently steer specific words using [Pronounced …]:

You can sprinkle these throughout the first verse; the model “locks in” the vowel set early.

✅ Step 1: In Genre field →

✅ Step 2: In your main prompt or lyric header →

✅ Step 3 (optional): Nudge tricky words phonetically →

That’s the current top-tier combo.

Example: Type up to 5 genres or vibes field: Australian, folk, acoustic, storytelling, heartfelt

🎵 Result: Natural Aussie inflection, local rhythm, and correct phrasing.

Include one or more of these in your Type up to 5 genres or vibes field or at the top of your lyrics block:

Or in the short 100-char genre field:

Smooth R&B voice, urban accent, soulful tone, no country twang

You can safely use accent cues without referencing artists:

or:

Rap, rhythmic spoken flow, East-Coast accent, no twang

This tells the model to switch to a different vocal profile:

Combine with the “no twang” hint if needed.

Add these exact phrases:

The “no melody” part stops it from defaulting to a sung country lilt.

If you really need to pull the model away from that country timbre, stack the genres in Type up to 5 genres or vibes field: R&B, soul, gospel, blues, urban pop

The more “urban” or “soulful” tags you include, the more the AI shifts tone color.

Allowed descriptors include: [Pronounced …], [Latin pronunciation], [Italian pronunciation], [Clear vowels], [Choral phrasing]

You can gently cue natural hesitations or breath pauses without disrupting flow:

Example:

Or inline:

These are compliant and help tone, not mimic artists: