How modern AI listens, learns, and speaks back — in your voice.
Voice and AI is a plain-English deep dive into the technology that powers today's voice cloning and neural text-to-speech systems. Starting from how a microphone captures sound, the book walks through spectrograms, vocoders, transformer-based speech models, and the ethics of synthetic voices. It's the same playbook we used to build Clone Voice Translator — written so you can read it on a flight and walk off understanding the field.
Sampling · Spectrograms · Embeddings · Vocoders · Diffusion · TTS
The chapters move from the physics of sound to production AI systems. Each one is short enough to read in a sitting and ends with a checklist of what you should now understand.
Pressure waves, sampling rates, and why 16 kHz audio is enough for almost everything voice AI does.
How models turn raw audio into mel-spectrograms — the picture-of-sound that neural networks actually look at.
The "fingerprint" vectors that let an AI capture you from just a few seconds of audio.
Tacotron, FastSpeech, VITS, and today's diffusion- and transformer-based TTS models — what each gets right and wrong.
How HiFi-GAN and neural vocoders rebuild audio from spectrograms, and how zero-shot cloning actually works.
Watermarking, identity verification, and the rules a responsible voice-AI product has to live by.
Chapter companions, behind-the-scenes engineering notes, and answers to the questions readers send us most often. New posts every few weeks.
Voice and AI, Chapter 5: why raw waveforms hide structure, how the Fourier Transform and short-time analysis reveal frequency over time, and why the spectrogram is what many models actually consume.