TTS Glossary
A quick reference guide to common TTS terminology, updated regularly.
TTS
Text-to-Speech: Technology that converts written text into natural-sounding speech.
Neural TTS
Neural TTS: Speech synthesis based on deep neural networks (e.g., Tacotron, FastSpeech, VITS), producing more natural quality than traditional concatenative synthesis.
Vocoder
Vocoder: A model that converts acoustic features (e.g., mel-spectrograms) into raw audio waveforms, such as HiFi-GAN, WaveGlow, WaveNet.
SSML
SSML: Speech Synthesis Markup Language for controlling prosody, pitch, pauses, emphasis and other synthesis parameters.
Phoneme
Phoneme: The smallest unit of sound in a language. TTS systems typically convert text to phoneme sequences before synthesis.
Voice Cloning
Voice Cloning: Learning a target speaker's voice characteristics from a small number of samples to generate new speech in that voice.
Zero-shot TTS
Zero-shot TTS: Synthesizing new voices without fine-tuning, relying on the model's generalization capability.
Few-shot TTS
Few-shot TTS: Voice cloning with seconds to minutes of reference audio.
Mel-Spectrogram
Mel-Spectrogram: A time-frequency representation of audio mapped to the human ear's perceptual frequency scale, commonly used as an intermediate feature in TTS.
VITS
VITS: Variational Inference Text-to-Speech, an end-to-end neural TTS model that produces high-quality speech directly from text.
Diffusion TTS
Diffusion TTS: Speech synthesis using diffusion models that generate high-quality audio through progressive denoising, e.g., NaturalSpeech series.
Streaming TTS
Streaming TTS: Synthesizing and outputting speech simultaneously, suitable for real-time dialogue and interactive scenarios with low latency.
Speaker Embedding
Speaker Embedding: A vector representation of a speaker's voice characteristics, used for voice control in multi-speaker TTS.
MOS Score
MOS: Mean Opinion Score, a subjective quality evaluation metric where listeners rate naturalness on a 1-5 scale.
Voice Converter / VC
Voice Conversion (VC): Changing a speaker's voice characteristics while preserving linguistic content, complementary to TTS.
End-to-End TTS
End-to-End TTS: Direct text-to-audio generation without intermediate steps like alignment or duration prediction, e.g., VITS, FastSpeech.
Chinese TTS
Chinese TTS: Chinese TTS requires handling polyphonic characters, prosodic boundaries, and tones. Recent models like ChatTTS and CosyVoice have shown outstanding results.
Singing Voice Synthesis / SVS
SVS: Singing Voice Synthesis, a specialized TTS application requiring simultaneous control of pitch, rhythm, and pronunciation.