A Study on How Human Annotations Benefit the TTS Voice

When we built the unit inventory from the Blizzard corpus, three types of manual works were performed. All these works took about 12 working days of our labelers. In order to see how much benefit these manual works bring us, we performed several perceptual experiments to compare the speech generated with/without manual works. The results show that although the manual proofreading identified more than 500 word-errors, no improvement is observed in our experiment. Both manual checking of segmental boundaries and manual prosody annotations make the synthesized speech better. And the later one brings more benefit. The preference rate between the final version of the synthetic speech with limited manual works and the fully automatically processed version is 68% to 32%.

• Publication year

  2006

• Venue

  Blizzard Challenge

• Publication date

  2006-09-16

• Fields of study

  Linguistics, Computer Science

• Identifiers
  DOI 10.21437/blizzard.2006-10
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Context-Dependent Boundary Model for Refining Boundaries Segmentation of TTS Units

  2006cited by this paper
• Modeling stylized invariance and local variability of prosody in text-to-speech synthesis

  2006cited by this paper
• Automatic accent annotation with limited manually labeled data

  2006cited by this paper
• Towards phone segmentation for concatenative speech synthesis

  2004cited by this paper
• Evaluating and correcting phoneme segmentation for unit selection synthesis

  2003cited by this paper
• Microsoft Mulan - a bilingual TTS system

  2003cited by this paper
• Automatic phonetic segmentation

  2003cited by this paper
• Perceptually based automatic prosody labeling and prosodically enriched unit selection improve concatenative text-to-speech synthesis

  2000cited by this paper
• Inter-transcriber reliability of toBI prosodic labeling

  2000cited by this paper
• Automatic ToBI prediction and alignment to speed manual labeling of prosody

  1999cited by this paper

• Emilia: a speech corpus for Argentine Spanish text to speech synthesis

  2019cites this paper
• On the impact of phoneme alignment in DNN-based speech synthesis

  2016cites this paper
• Multitier Annotation of Urdu Speech Corpus

  2014cites this paper
• Measuring a decade of progress in Text-to-Speech

  2014cites this paper