Scribd
Upload
255 views3 pages

Advanced Prompting For Expressive TTS

Eleven Labs' v3 model enhances text-to-speech capabilities by enabling expressive features like whispers, laughter, and emotional depth through careful voice selection, stability settings, and prompt structuring. Best practices include using pre-tuned voices, maintaining a prompt length of 250-1,000 characters, and embedding audio tags for emotion and delivery. For multi-speaker dialogue, it's essential to label each speaker and choose distinct voices to ensure fluidity in conversation.

Uploaded by

jayesh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
255 views3 pages

Advanced Prompting For Expressive TTS

Eleven Labs' v3 model enhances text-to-speech capabilities by enabling expressive features like whispers, laughter, and emotional depth through careful voice selection, stability settings, and prompt structuring. Best practices include using pre-tuned voices, maintaining a prompt length of 250-1,000 characters, and embedding audio tags for emotion and delivery. For multi-speaker dialogue, it's essential to label each speaker and choose distinct voices to ensure fluidity in conversation.

Uploaded by

jayesh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Section X: Eleven Labs V3 — Advanced Prompting for Expressive TTS

Overview:​
Eleven Labs’ v3 model elevates text-to-speech from “just speaking” to fully
performing—whispers, laughter, accents, multi-speaker dialogue, and nuanced emotional
depth. To harness it, you’ll combine the right voice, stability, and audio tags in your
prompts.

1. Voice Selection

●​ Choose a “Best Voices for V3” entry ( ✔️). These are pre-tuned for maximum
expressiveness.​

●​ Custom voices (your own recordings) require fine-tuning before v3 can apply
advanced emotional tags.​

2. Stability Setting

Controls adherence vs. creativity:

●​ Creative → Most emotional & flexible (↑ hallucination risk).​

●​ Natural → Balanced, closest to recorded style.​

●​ Robust → Most stable, least responsive to tags.​

Tip: For tag-driven emotion, start at Natural or Creative; for repeatable


automation, slide toward Robust.

3. Prompt Structure & Length

●​ 250–1,000 chars yields the most consistent, expressive output.​

●​ Ellipses (…) introduce natural pauses.​

●​ Em-dashes (—) or parentheses add emphasis or side comments.​

●​ Capitalization highlights key words: e.g. “WOW,” “UNBELIEVABLE.”​

4. Audio Tags

Embed inline to dictate emotion, delivery, and non-verbal cues. Use before or mid-sentence:
●​ Voice Cues:​

○​ Whisper: “…”​

○​ Laughs: | Laughs harder: | Chuckles:​

○​ Sigh: | Exhale:​

○​ Sarcastic: | Curious: | Excited:​

●​ Sound Effects:​

○​ Applause: | Gunshot: | Footsteps: | [Music]​

●​ Accent Tags:​

○​ Strong French accent: | Mild Australian accent: | etc.​

●​ Multi-Speaker:​

○​ Speaker A: / Speaker B: to alternate voices in the same clip.​

Example Prompt:

vbnet
CopyEdit
Whisper: “Hey, Jessica…”
Laughs: “That is absolutely incredible!”
Strong Australian accent: “Fancy a cup of tea?”
Applause: “Thank you, everyone!”

5. Multi-Speaker Dialogue

●​ Prefix each line: A: / B: / C:​

●​ Choose different V3 voices for each speaker.​

●​ Maintain Natural or Creative stability for fluid back-and-forth.​

Example:

vbnet
CopyEdit
A: Sigh: “I may have tried to debug myself….”
B: Sarcastic: “Oh wow, you really did it this time.”
A: Curious: “Why does my voice keep glitching?”

Best Practices Recap:

1.​ Voice: Pick a V3-verified voice.​

2.​ Stability: Start at Natural/Creative; adjust toward Robust for repeatability.​

3.​ Prompt Length: ≥250 characters for nuance.​

4.​ Tags: Sprinkle emotional, delivery, and sound-effect tags inline.​

5.​ Punctuation: Use ellipses and em-dashes for natural rhythm.​

6.​ Multispeaker: Label each speaker and assign separate voices.

You might also like