Vowel estimation accuracy varies by pronunciation order and duration: Creating TTS evaluation data and two measurement pitfalls
๐ Originally published (in Japanese) at forge.workstyle.tech . Target Audience : Developers implementing voice analysis or avatar lip-syncing in browsers. Input and Output : Labeled TTS audio is input into a vowel estimator, and the accuracy rate is obtained from the correct answers and estimated results. What You'll Gain from This Article : The reason why the order and length of audio affectโฆ
The accuracy of vowel estimation in Text-to-Speech (TTS) systems can vary depending on the pronunciation order and duration of the audio. This is because stateful estimators update their internal representation based on the audio features received. In this study, researchers found that the order and length of audio had a significant impact on the evaluation results of a vowel estimator.
They discovered that the measurement method itself could create the answer, rather than accurately assessing the system's performance. To address these issues, they synthesized sustained vowels like "ใใผใผใผ" and "ใใผใผใผ" using a Style-Bert-VITS2 model and verified the conditions with leave-one-speaker-out analysis. Additionally, they randomized the order of vowels and reduced their length to 120ms to minimize the influence of the long-term average on the estimation results.
Written by urgent.news from Dev.to's reporting โ not their text. Machine-written โ may contain errors; check the original before relying on it.
