Text-to-Speech

Text-to-Speech, usually shortened to TTS, is technology that converts written text into spoken audio. Early systems stitched together recorded fragments and sounded mechanical; current systems use neural models that generate the waveform directly, learning prosody, pauses, and emphasis from large speech datasets, which is why modern output carries intonation rather than a flat monotone. Most services expose a set of preset voices plus voice cloning, where a short sample is enough to produce a synthetic voice that resembles a specific speaker. ElevenLabs built its business on this quality, OpenAI and Google offer speech synthesis through their APIs, and Amazon Polly and Microsoft Azure serve enterprise workloads. The uses are practical: audiobook narration, accessibility for readers with visual impairments, voiceover for video, and spoken replies from AI assistants. Latency matters as much as fidelity for conversational products, since a natural exchange breaks down if the reply takes too long to start. The pitfall is consent and misuse. Cloning a voice without permission enables impersonation and fraud, so major vendors gate cloning behind verification and embed audio watermarks. TTS is the inverse of speech recognition, and voice assistants chain the two around a language model.