8.8.6.1.3 - How to Use ElevenLabs: Speech-to-Speech for Emotion Transfer (Difficulty: Advanced | Path: Scale)

Lesson Summary

Mastering Nuance with Speech-to-Speech (STS)

What is it? Text-to-Speech (TTS) guesses how a sentence should be read based on punctuation. Speech-to-Speech (STS) allows you to upload an audio file of you speaking, and the AI will replace your voice with the target voice while keeping your emotion, speed, and intonation.

Why It’s a Game-Changer

Sometimes, TTS just doesn't \"get it.\" It might read a sarcastic joke seriously, or sound bored during an exciting announcement. STS solves this by letting you \"act out\" the script. You don't need a good voice; you just need the right energy. The AI fixes the timbre/tone, but keeps your performance.

Step-by-Step Workflow

1. Record the Reference: Record yourself reading the script on your phone or computer. Don't worry about sound quality or your voice pitch. Focus entirely on the pacing, pauses, and energy. If you want the AI to whisper, you whisper. If you want it to shout, you shout.
2. Upload to ElevenLabs: Go to 'Speech Synthesis' and switch the tab from 'Text to Speech' to 'Speech to Speech'. Upload your audio file.
3. Choose the Target Voice: Select the professional voice you want to use.
4. Adjust Settings: Set 'Style Exaggeration' to standard. If the output sounds too much like you and not enough like the AI model, increase the stability settings.
5. Generate: The AI will fuse your performance with the professional voice skin.

Do’s and Don’ts

• Do: Use STS for high-stakes lines, like the hook of an ad or a punchline, where timing is everything.
• Don't: Upload a reference file with loud background noise or music. The AI tries to interpret the noise as speech, resulting in weird artifacts. Keep the recording clean.
• Do: Over-act slightly in your reference recording. AI models sometimes \"dampen\" emotion, so giving 110% energy often results in a perfect 100% output.

Real-Life Example: A founder wanted to use a \"deep movie trailer voice\" for a Halloween sale ad. When they used Text-to-Speech, it sounded flat. They switched to Speech-to-Speech, recorded themselves whispering menacingly into their iPhone, and applied the deep voice skin. The result was a terrifyingly realistic, high-production trailer that went viral for its creativity.