Voice cloning generates speech that sounds like a specific person using samples of their voice. Approaches: Speaker adaptation: Fine-tune TTS model on target speaker samples (minutes to hours of audio). Zero-shot cloning: Encode speaker embedding from few seconds of audio, condition TTS on embedding. Speaker verification: Confirm cloned voice matches target identity. Key models: YourTTS, XTTS (Coqui), Tortoise TTS, VALL-E (Microsoft, 3 seconds), ElevenLabs (commercial leader). Technical components: Speaker encoder (extract voice characteristics), speaker-conditioned decoder, voice consistency across utterances. Quality factors: Voice similarity, naturalness, expressiveness preservation, accent/language handling. Ethical concerns: Deepfakes for fraud, impersonation, misinformation, non-consensual voice use. Many platforms require consent verification. Detection: Watermarking cloned speech, deepfake detection models, provenance tracking. Legitimate uses: Deceased voice preservation, accessibility, personalized assistants, dubbing, voice banking for those losing speech. Balance between innovation and harm prevention is critical.