Short Description

Automatic Speech Recognition (ASR) is a technology that finally gets widely used in end-user products. While one approach is to only do the recording and feature extraction on the end-user device and do the processing in the cloud (e.g. Google ASR), for many applications we do not want to send audio recordings into the cloud. To make ASR accessible for many applications running the system on an embedded computer would be desirable. The Raspberry Pi is such a low-cost embedded computer, where we would like to run an ASR system for Austrian German. An open-source ASR engine for embedded computers is available, but no acoustic models exist for Austrian German.

Your Tasks

• Installation of CMU pocketsphinx on a Raspberry Pi
• Training and adaptation of acoustic models for Austrian German
• Transfer of an Austrian German Pronunciation Dictionary into the pocketsphinx format
• Design of a deterministic grammar for command and control tasks
• Optional if done as a team: Implement a simple voice dialog system
• Documentation

Your Profile

• Motivation and interest in the topic
• Background in Linux

Contact

Martin Hagmüller (hagmueller@tugraz.at or 0316/873 4377)