A smart eavesdropping system : recognizing keywords by human subjects
Speech Recognition (SR) gains its popularity in research area as the advance of modern technologies. It can translate speech into text with the aid of computers and speech recognition applications. In this final year project, an open source speech recognition engine named Pocketsphinx from Carnegie...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
2015
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/63768 |
| Summary: | Speech Recognition (SR) gains its popularity in research area as the advance of modern technologies. It can translate speech into text with the aid of computers and speech recognition applications. In this final year project, an open source speech recognition engine named Pocketsphinx from Carnegie Mellon University (CMU) is integrated into our existing Eavesdropping System to perform speech recognition or keyword spotting tasks on the output audio files from the system. This report covers the details on the development of whole speech recognition framework assembled. Pocketsphinx is compiled and installed in a Linux server remotely and communicates with the client Matlab programs using network sockets. System parameters are carefully tuned to ensure the performance. Experiments on different combinations of Acoustic Models (AM) and Language Models (LM) are also conducted and evaluated. Acoustic model adaptation which adapts the speech recognizer into specific acoustic environment or speaker to enhance the recognition performance is also presented. Furthermore, another commercially available speech recognition application named Dragon Naturally Speaking (DNS) 12 is also experimented and compared with Pocketsphinx used in the system. |
|---|