Scholarlink

<p>Voice casting has always been challenging in the multimedia industry. Recent research shows that voice casting can be done with the help of speaker recognition methods. In this paper, the first dataset for dubbing from English to Persian (Farsi) is introduced named E2PCast, English to Persian voice casting dataset. The E2PCast dataset includes audio segments from 30 movie characters (15 females and 15 males) in Persian and English. Approximately, the dataset contains a total of 2.5&#xa0;h of audio in Persian and English languages (5&#xa0;h and 17&#xa0;min in total). In addition to explaining the description of data collection, this paper reports the results of tests performed on the dataset. The experiments and their evaluations include the automatic voice casting benchmark system and speaker identification tests to check characters’ overall identity distribution in the dataset. Based on the benchmark results, the overall test accuracy of the Siamese neural network method on the dataset is 63.52%. Also, the average accuracy of the 5-fold cross-validation test (without common identity in the train and development data) is equal to 52.58%. The E2PCast dataset will be available for free for educational purposes only.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

E2PCast: an English to Persian voice casting dataset

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • Sogol Alipour Esgandani,
  • Yasser Shekofteh

摘要

Voice casting has always been challenging in the multimedia industry. Recent research shows that voice casting can be done with the help of speaker recognition methods. In this paper, the first dataset for dubbing from English to Persian (Farsi) is introduced named E2PCast, English to Persian voice casting dataset. The E2PCast dataset includes audio segments from 30 movie characters (15 females and 15 males) in Persian and English. Approximately, the dataset contains a total of 2.5 h of audio in Persian and English languages (5 h and 17 min in total). In addition to explaining the description of data collection, this paper reports the results of tests performed on the dataset. The experiments and their evaluations include the automatic voice casting benchmark system and speaker identification tests to check characters’ overall identity distribution in the dataset. Based on the benchmark results, the overall test accuracy of the Siamese neural network method on the dataset is 63.52%. Also, the average accuracy of the 5-fold cross-validation test (without common identity in the train and development data) is equal to 52.58%. The E2PCast dataset will be available for free for educational purposes only.