Signal Processing and Speech Communication Laboratory

Welcome!

In 2000, the Signal Processing and Speech Communication Laboratory (SPSC Lab) of Graz University of Technology (TU Graz) was founded as a research and education center in nonlinear signal processing and computational intelligence, algorithm engineering, as well as circuits & systems modeling and design. It covers applications in wireless communications, speech/audio communication, and telecommunications.

If you want to learn more about Signal Processing, click: What is Signal Processing?

The Research of SPSC Lab addresses fundamental and applied research problems in five scientific areas:

Result of the Month

On Disfluency and Non-lexical Sound Labeling for End-to-end Automatic Speech Recognition [link]

Abstract: Spontaneous speech contains a significant amount of disfluencies and non-lexical sounds (e.g., backchannels, filled pauses), which are often difficult to transcribe. Disfluency labeling for automatic speech recognition (ASR) aims at editing these phenomena in the transcription to improve overall recognition accuracy. Such labeling techniques typically delete nonlexical/disfluent labels from the prediction, where classical ASR techniques either ignore or treat them as lexical items. Our results, obtained by systematic comparison and detailed evaluation of various disfluency labeling methods on two different language conversational corpora, suggest that neither of the previous approaches are optimal. We propose to distinguish between filled pauses and meaningful conversational grunts and show that keeping the non-lexical labels is not only possible but as low as 7% label error rates can be achieved for highly important categories (including ’mhm’) while preserving a decent WER.

Index Terms: end-to-end speech recognition, disfluency, conversational speech, filled pauses, Hungarian, Austrian German

Konferenz: Proc. of Interspeech 2024, pp. 1270–1274, 2024. (1-5 September 2024, Kos, Greece)

Paper-Link (pdf)

Contact: Julian Link, Barbara Schuppler