Text-to-speech takes text and produces an audio signal that speaks the words. A system must determine pronunciation, phrasing, rhythm and emphasis before a decoder or vocoder renders the final waveform.
Modern systems may predict acoustic features, discrete codec tokens or continuous latent representations. The chosen representation affects naturalness, control, latency and computational cost.
Text-to-speech is used for accessibility, narration, customer support and voice agents. Responsible deployments also consider consent, speaker identity, disclosure and safeguards against deceptive impersonation.
ELI5
Text-to-speech is software that reads written words aloud. It decides not only which sounds to make, but also where to pause and which words to emphasize.
For example, a navigation app can turn the written instruction 'turn left' into a spoken direction. A good system makes the result clear and natural instead of sounding like disconnected syllables.
Is text-to-speech the same as speech recognition?
No. Text-to-speech turns text into audio, while speech recognition turns audio into text.
What controls the quality of generated speech?
Training data, pronunciation handling, prosody modeling, the audio representation and the final decoder all affect quality.
