본문으로 건너뛰기
AI 도감

음성 합성

글을 자연스러운 사람 목소리로 읽는다

음성과 음악입문 #28
입력텍스트오디오

이 페이지의 본문은 영어로 제공됩니다. 제목과 요약은 한국어로 번역되었습니다.

이 능력이 뜻하는 것

Takes text and outputs audio that reads it aloud. It must handle both pronunciation — saying the right words — and prosody — where to pause and what to stress — while carrying a target speaker timbre. It runs opposite to speech recognition, and unlike voice cloning it does not need to imitate a specific person from a few samples but usually picks from preset voices.

기술적으로 구현하는 방법

Modern systems are mostly two-stage: an acoustic model predicts an intermediate representation such as a mel spectrogram from text, then a vocoder restores the waveform; both can be trained end to end. Discrete audio tokens with autoregressive or diffusion decoding yield more natural prosody and support zero-shot timbre imitation. On the text side, phonemisation and text normalisation handle digits, abbreviations and polyphonic characters.

대표 제품

6

관련 기관

대표적 용도

  • Audiobooks and news reading
  • Accessibility reading and assistive narration
  • Navigation prompts, podcasts and dubbing
  • Support and smart-device response audio

성능을 평가하는 방법

MOS
Mean opinion score for naturalness
Intelligibility (WER)
Re-transcribing the output to measure how much is misheard
Speaker similarity
Closeness to the target timbre

경계와 난점

  • Pauses and stress in long sentences sound unnatural, making whole passages feel flat
  • Numbers, proper nouns and polyphonic characters are misread, and one normalisation slip derails the sentence
  • Control over emotion and style is coarse; only a few systems let you finely set pace and mood

뒤에 있는 개념