IBM - CVUT Student Research Projects Web To Speech (Text To MP3) Josef Kejzlar (kejzlj1@fel.cvut.cz) Petr Vavroušek (vavrop1@fel.cvut.cz) Vojtěch Hora (horav1@fel.cvut.cz) Martin Hevera (heverm1@fel.cvut.cz)
How It Works 1) Client: User enters the website using Firefox with WebToSpeech plugin installed. 2) Client: Click “process page” or “process selection” and send requested page (part of page) to server. 3) Server: Parser extracts text from html source using template matching given URL or generic algorythm if no template is found. 4) Text is transformed into speech using TTS, encoded into MP3s, compressed together with metadata file and sent back to the client. 5) Client: Received files is transferred into handheld device. 6) Handheld: Flash application locates new files and allows user to play the content of web page. IBM - CVUT Student Research Projects 2
Overview • We continue in project created last year. See the previous presentation. • Main tasks: 1) Complete server-side application 2) Complete FireFox plugin 3) Port flash viewer to mobile platform (Flash Lite) 4) Improve file format 5) Improve parsing (extracting text from html page) IBM - CVUT Student Research Projects 3
1) Complete server-side application Communication Protocol It was necessary to implement some mechanism for invoking server-side process from the client-side Firefox plugin. HTTP protocol was chosen, because it has possibility to pass parameters and return multiple results (using HTTP push method). This allows client to see the progress of request processing. Text To Speech engine Because the already implemented TTS engine from IBM is not freely available, for videly usage it is necessary to implement also some free TTS. The Festival engine has been chosen. IBM - CVUT Student Research Projects 4
2) Complete Firefox plugin Previous Firefox plugin was not suitable for major code redesign, and therefore completely new plugin has been writen. User Interface New UI has two entries in right-click popup menu. 1) Process whole page 2) Process only selected text region 2 Basic dialogs: 1) Main processing dialog 2) Settings IBM - CVUT Student Research Projects 5
Main processing dialog Settings dialog IBM - CVUT Student Research Projects 6
3) Port flash viewer to mobile platform There are some issues related to multithreading and thread synchronisation in Flash lite, therefore the application is not yet fully usable on mobile devices. Also the GUI has to be redesigned to fit the handhelds screen size and limited number of device controls. (work in progress) IBM - CVUT Student Research Projects 7
4) Improve file format Previous format had all the information in ID3 tag. In the new version there was request to store additional information (like hyperlinks) and therefore it was necessary to choose new data storage. Old format also contains no information about articles count, articles names, source url etc. New data storage XML file has been introduced. It contains all the metadata stored previously in ID3 tags plus some additional information. (Work in progress. XML format has been specified, waiting for implementation) IBM - CVUT Student Research Projects 8
5) Improve parsing Extracting important textual information from html web page is quite dificult task. Write generic parser that would work for every web page is impossible. Therefore the server and probably even the Firefox plugin will use “templates” specialy written for the most common pages. These XML templates will tell the parser where to find the requested text in HTML DOM structure. (Work in progress. Fully designed (including algorythms), not yet implemented) IBM - CVUT Student Research Projects 9
Recommend
More recommend
