pb425 - Modern Speech Technology (Complete module description)

Workload 28 hours + 62 self study

First Insight into Signal- or System Theory

The course provides the engineering tools for modern speech signal processing. It complements the physiological aspects of speech generation and the psychoacoustic aspects of speech perception provided by the companion course "Einführung in die Sprachverarbeitung" by teaching the technical representation of speech in algorithms, hardware and software. Special attention is devoted to the task of speech signal enhancement in several facets, such as noise filtering, reverberation reduction and echo cancellation. The required tools of modern speech processing are presented here in an elementary and intuitive manner. Mathematical requirements are moderate to low. The exercises are implemented in software or on the blackboard by the students themselves or in teams and, if necessary, with component-wise support by the lecturers. The students thus gain a preliminary insight into the scientific working method for student qualification projects, such as the bachelor thesis, as well as into the engineering working method in the industrial environment of speech processing.

1.         Technical Representation of Speech Signals
- Bandwidth, sampling rate, sampling rate error, bit depth
- Single-channel, multi-channel, and binaural signals
- Acoustic sensor network
- Databases for speech and noise signals
- Acoustic room simulation
 

2.         Speech Signal Enhancement Tasks
- Competitive speech enhancement challenges
- Tasks with ambient noise, reverberation, or interferences
- Evaluation metrics: PESQ, STOI, SegSNR, POLQA, ViSQOL
- Model-based processing paradigm: e.g., Wiener-Filter
- DNN-based processing paradigm: i.e., end-to-end with data
 

3.         Hardware- und Software-tools for speech processing
- Integrated development envs: Matlab, Python, TensorFlow
- Commercial solutions: e.g., Nvidia, Krisp
- Devices for speech acquisition and processing: Personal Computer, Mobile, Raspberry Pi, Hearing Aid
 

4.         Remote Speech Communication
- Systems for speech communication: fixed telephone network, mobile network, Voice-over-IP,
   conferencing systems, NFMI
- Fundamentals of transmission: source and channel coding
- Emerging services: WebRTC, Speex, EVS
- Principles and properties of codecs: Sampling rate, bitrate, latency,
- Packet loss: statistical modeling and concealment

Vary, Martin: Digital Speech Transmission – Enhancement, Coding, and Error Concealment, Wiley 2006

1 Klausur (30 - 60 Min.) oder
1 mündliche Prüfung (20 - 30 Min.) oder
1 Referat (20 -30 Min.)