TTS Glossary

The rabbit-hole engine. Every technical term used on this site has a full definition with examples, related concepts, and links to the relevant reviews and guides.

Pricing

Character economics

Character economics is the cost-per-million-characters analysis that reveals which TTS vendors are cheapest at real production volumes — and why ElevenLabs' pricing model punishes scale.

pricingcostscale

Performance

First-byte latency

First-byte latency is the time from sending a TTS request to receiving the first audio byte — the metric that determines whether a tool is usable for real-time AI agents, IVR, or live captioning.

latencystreamingreal-time

Voice Quality

MOS (Mean Opinion Score)

MOS is the 1-to-5 subjective voice quality scale that TTS vendors still use — and why it no longer differentiates the leading tools in 2026.

qualitybenchmarkmeasurement

Voice / Audio Technology

Neural TTS

Neural TTS uses deep learning models to synthesize human-like speech from text — replacing the old sample-stitching approach and enabling the realistic AI voices of 2026.

neural TTSWaveNetdeep learning

Prosody

Prosody is the rhythm, stress, intonation, and timing of speech — the difference between robotic and human-sounding voice. The primary dimension on which modern TTS tools still vary.

prosodyintonationrhythm

SSML

Speech Synthesis Markup Language — XML-based markup that controls how TTS engines pronounce text, add pauses, adjust pitch, and handle emphasis.

SSMLSpeech Synthesis Markup Languageprosody

Voice Cloning

Voice cloning uses AI to replicate a specific person's voice from sample audio. Requires explicit consent — cloning without consent is illegal in most jurisdictions.

voice cloningAI voiceElevenLabs

Voice Model

A voice model is the trained neural network that defines how a TTS system sounds — each vendor's flagship 'voice' is a model version with specific quality, latency, and language tradeoffs.

voice modelTTS modelAI voice