Level 6Lesson 52⏱️ 40 min

AI Voice & Audio

Text-to-speech, voice cloning, music generation, and dubbing - with the consent and licensing rules that come with them

Four Things AI Audio Can Do

Audio generation splits into four jobs, each with different leading tools and different ethical weight:

The audio toolkit:
  • Text-to-speech (TTS) - turn a script into natural narration.
  • Voice cloning - recreate a specific voice from a sample.
  • Music generation - full songs or instrumental beds from a prompt.
  • Dubbing & transcription - translate/lip-sync, or turn speech into text.

Text-to-Speech: Narration Without a Mic

Modern TTS captures breath, emphasis, and emotion well enough for explainer videos, audiobooks, IVR systems, and accessibility. ElevenLabs is the 2026 quality leader; Google and OpenAI also ship strong voices.

What you control in good TTS:
  • Voice selection (gender, age, accent, tone)
  • Pacing and pauses (often via punctuation or tags)
  • Emphasis and emotion ("excited", "calm", "serious")
  • Multi-speaker dialogue (assign different voices)

Good for: explainers, e-learning, audiobooks, prototypes,
          accessibility narration, podcast drafts.
Quality tip: Write for the ear. Short sentences, natural contractions, and deliberate punctuation produce far better speech than dense written prose.

Voice Cloning: Powerful and Sensitive

Cloning recreates a specific person's voice from a short sample. Legitimate uses are real - narrators scaling their own voice, dubbing an actor across languages, accessibility for people losing their speech. So is the potential for abuse.

The consent rule - non-negotiable: Only clone a voice you own or have explicit, documented permission to use. Cloning someone's voice without consent is unethical and, in a growing number of places, illegal. Reputable tools require voice verification for exactly this reason.

Music Generation

AI can now produce full songs with vocals, or instrumental beds for video. The tool choice is a direct quality-vs-licensing trade-off - exactly the tension from Lesson 48.

Suno v5           Best overall song quality, vocals, genre range.
                  ⚠ Commercial license terms unsettled
                    (training-data lawsuits in flight, 2026).

ElevenLabs Music  Trained on LICENSED data → clean commercial terms.
                  Quality very strong, sometimes a step behind Suno.

Udio              Strong quality; similar licensing caution as Suno.

Google Lyria      Instrumental + real-time streaming options.
For anything you'll publish or monetize: favor a tool with clear, licensed commercial terms (ElevenLabs Music, Stable Audio) over the highest-quality tool with murky rights. A great track you can't legally use is worthless.

Dubbing & Transcription

Dubbing

Translate a video into another language in the original speaker's voice, with lip-sync. ElevenLabs and others make a single video instantly multilingual.

Transcription

Speech-to-text (Whisper-class models) for captions, subtitles, searchable archives, and meeting notes. Often free or near-free and highly accurate.

Accessibility win: Auto-captions and transcripts aren't just nice - they widen your audience and are increasingly expected (and sometimes required).

Hands-On: Narrate and Score a Clip

Hands-on (20 min): Write a 4-sentence script and generate it as voiceover in a TTS tool (ElevenLabs has a free tier). Try two different voices and notice how tone changes the message. Then generate a short instrumental bed in a music tool with clear commercial terms. You now have a voice track and a music track - the audio layer you'll pair with video in the capstone. Keep both files.
Lesson 52 Quick Reference
Text-to-speech (TTS)

Script → natural narration; ElevenLabs is the 2026 quality leader

Write for the ear

Short sentences, contractions, deliberate punctuation produce better speech

Voice cloning

Recreate a specific voice from a sample - only with documented consent

Suno v5

Best song quality with vocals, but commercial license terms are unsettled in 2026

ElevenLabs Music

Licensed training data → clean commercial terms; the safe choice for published work

Dubbing

Translate video into another language in the original voice with lip-sync

Transcription

Speech-to-text for captions, subtitles, and accessibility - often free and accurate

← AI Video Generation
Unlocks in ~10 min of reading