Texttospeech

Text-to-speech (TTS) is technology that converts written text into spoken language. A typical TTS system processes text through stages such as text normalization (expanding numbers and abbreviations), linguistic analysis (pronunciation, syntax, and prosody), and voice synthesis that generates audible speech. Some systems also include post-processing for pronunciation adjustments, style, and pacing. The result is an audio waveform delivered in real time or offline.

TTS methods vary. Concatenative synthesis stitches together prerecorded speech units to form natural-sounding utterances, with voice quality depending on unit selection and recording coverage. Parametric or statistical parametric synthesis models voice characteristics such as pitch, duration, and spectral features, often using a vocoder to reconstruct audio. Neural text-to-speech uses deep learning to predict acoustic features or waveform directly, typically offering higher naturalness and flexibility, with vocoders such as WaveNet, WaveGlow, or HiFi-GAN enabling high-quality output.

Applications include accessibility tools such as screen readers, navigation systems, virtual assistants, automated customer service, language learning, and media narration. TTS supports many languages and voices, enabling gendered or characterful personalities and on-device or cloud-based deployment. Licensing, data privacy, and on-device processing influence performance and user control.

Quality is evaluated by objective and subjective measures; common metrics include mean opinion score (MOS) for naturalness and intelligibility, as well as pitch and spectral error. Challenges include handling homographs, abbreviations, and emotional prosody, as well as dialectal variation and code-switching. Ethical considerations address bias, safety, consent for voice data, and potential misuse such as impersonation.

As a field, TTS continues to advance with larger datasets, more efficient models, multilingual capabilities, and customizable voices. It remains a core component of assistive technology and spoken-user interfaces.