Download presentation
Presentation is loading. Please wait.
Published byJohn McCormick Modified over 9 years ago
2
SPEECH SYNTHESIS --AusTalk Zhijie Shao Master of Computer Science Supervisor: Trent Lewis
3
Application http://en.wikipedia.org/wiki/File:Stephen_Hawking.StarChild.jpg http://www.popsci.com.au/technology/article/2009-12/secrets-behind- brain-implanted-speech-sythesizer-revealed-new-paper
4
Voice Import ToolBlizzard DataAusTalkModificationEvaluation and Conclusion Project Procedure
5
MARY (Modular Architecture for Research on speech sYnthesis) is the German text-to-speech system. MARY
6
Voice Import Tools being provided by MARY contains a set of Voice components and helps users to build new voices under the MARY Environment. MARY— Voice Import Tool
7
Unit Selection Synthesis
8
HMM- Based Synthesis
9
HMM- based Synthesis HMM: Hidden Markov model An introduction to HMM-based speech Synthesis by Junichi Yamagishi
10
Blizzard Data The speaker is known as ‘Nancy’ and is a native speaker of US English, professional female voice talent, voice coach, and singer. 16.6 hours of data was made available Unit Selection Voice: HMM-based Voice: Unit Selection Voice: Hi, I am Jacky, welcome to my presentation. Hi, Jacky again, welcome to my presentation.
11
AusTalk Data “7. Sentences (8mns) A set of 59 sentences is presented one at a time on the screen.” from BigASC-RA-Manual
12
Comparison Blizzard 59 VS AusTalk 59 Synthesis Blizzard Unit Selection: Blizzard HMM: AusTalk HMM: AusTalk Unit Selection:
13
Comparison Continue Blizzard Unit Selection: Blizzard HMM: AusTalk HMM: AusTalk Unit Selection: Welcome to the speech synthesis
14
Austalk vs Blizzard (1) Phoneme Alignment (2) Quality of wav files (3) Boundary of utterance Comparison
15
Evaluation and Conclusion Modification and Evaluation Create a quality Aussie voice Further Research
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.