| Speaker Name | Sachin Joshi |
|
| Organization | International Institute of Information Technology, Hyderabad | |
| Type | Talk | |
| Slides | Click to download | |
Building Tools using Hindi Speech Recognizer |
||
| Abstract | Recently, I build a Hindi Automatic Speech Recognition (ASR) system using CMU Sphinx, which is available at http://sourceforge.net/projects/hindiasr . This system is in the form of abstract acoustic models which can be plugged into any application program. The speech recognition technology, particularly in Indian languages, can extend accessibility to many illiterate masses, kids and visually handicapped people. Indian language ASRs will play a major role in localization of operating systems also. The mission clearly poses two distinct challenges: 1. Building acoustic models for indivisual languages 2. Building applications on top of it to make the ASR actually accessible to end user. In case of Hindi, we have already completed building acoustic model. Now, to make this system available to people is a communitywide task. N number of advanced applications can be developed using Hindi ASR. To name a few for example - driving menus and other GUI components using voice commands, dialog systems, dictation systems, games etc. The talk is oriented towards developer community. The purpose of this task is to explain steps in integration of HindiASR with any end user application. The talk will brief about what are the components of speech recognition system, then it will provide details about how to build a domain specific language model and finally it will explain important APIs provided by CMU Sphinx. The knowledge of these APIs is crucial for application developers. The talk may include a 5 minute of demo at the end. The audience will get benefited in following ways - they will come to know what is the current state of art of ASR technology and related opensource tools, they will understand whole procedure of using HindiASR in their system, they will know how to configure the system according to their domain specific needs. This will enable them to either write their own applications or to participate in HindiASR project itself, whose future objective is to build good ASR based applications for Indian society. |
|
| Pre-requisites | No specific prerequisite. But they should know C++ or Java to understand Sphinx APIs. | |
| Speaker Profile | I am Computer Science Engineering graduate from Shivaji University Maharashtra. Currently I am Ph.D. Participant at IIIT Hyderabad. I work in field of speech recognition. I have worked on 3-4 projects for building Multispeaker Large Vocabulary Continuous Speech Recogntion systems especially for Indian Languages. Recently I built Acoustic Models for Hindi using CMU Sphinx toolkit which are very useful for application developers in FOSS community. (Ref: http://sourceforge.net/projects/hindiasr/) This work was done under FLOSS 2007 fellowship given by Sarai. I have 5 International Publications. | |
Comments:
No comments on this talk yet.
