“Hulan-2013-12-27”版本间的差异
来自cslt Wiki
(以内容“=ASR= ==ASR Kernel development== http://cslt.riit.tsinghua.edu.cn/mediawiki/index.php/2013-12-20 ASR group weekly report ==TTS== * This week :* Huilan female v...”创建新页面) |
|||
第8行: | 第8行: | ||
* This week | * This week | ||
− | :* | + | :* Deliver a female voice |
− | :* | + | :* Deliver a tunable pitch/speaking rate |
− | :* | + | :* Male voice labelled finished |
− | + | :* Pronunciation dictionary check | |
− | + | ||
− | + | ||
− | :* | + | |
=Dialog system= | =Dialog system= | ||
第20行: | 第17行: | ||
==Statistical approach== | ==Statistical approach== | ||
− | * | + | * A preliminary book-search demo is ready, with 400 books, 6 properties, 10 patterns. |
− | + | * Looks fine, while the pattern set is still limited | |
− | + | * Need to expand the patterns and properties and enrich the functionality. | |
− | + | * Looking for more efficient pattern match/learning approach. | |
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
− | + | ||
==Template matching== | ==Template matching== | ||
− | * New words | + | * New words extraction |
− | * | + | :* Extracting person names |
+ | :* Rule-based extraction, consider surname, name convention | ||
+ | :* N-gram filter out non-names | ||
+ | :* Active word embedded composed words filtered out | ||
+ | :* Check possible names by search engine | ||
=System design= | =System design= | ||
− | * Interface | + | * Interface implementation on going (dimension management). |
− | * | + | * Deadline is 14/1/17 |
2013年12月27日 (五) 01:22的版本
目录
ASR
ASR Kernel development
TTS
- This week
- Deliver a female voice
- Deliver a tunable pitch/speaking rate
- Male voice labelled finished
- Pronunciation dictionary check
Dialog system
Statistical approach
- A preliminary book-search demo is ready, with 400 books, 6 properties, 10 patterns.
- Looks fine, while the pattern set is still limited
- Need to expand the patterns and properties and enrich the functionality.
- Looking for more efficient pattern match/learning approach.
Template matching
- New words extraction
- Extracting person names
- Rule-based extraction, consider surname, name convention
- N-gram filter out non-names
- Active word embedded composed words filtered out
- Check possible names by search engine
System design
- Interface implementation on going (dimension management).
- Deadline is 14/1/17