Florian Eyben

Andreas Triantafyllopoulos, Johannes Wagner, Hagen Wierstorf et al.

Large, pre-trained neural networks consisting of self-attention layers (transformers) have recently achieved state-of-the-art results on several speech emotion recognition (SER) datasets. These models are typically pre-trained in self-supervised manner with the goal to improve automatic speech recognition performance -- and thus, to understand linguistic information. In this work, we investigate the extent in which this information is exploited during SER fine-tuning. Using a reproducible methodology based on open-source tools, we synthesise prosodically neutral speech utterances while varying the sentiment of the text. Valence predictions of the transformer model are very reactive to positive and negative sentiment content, as well as negations, but not to intensifiers or reducers, while none of those linguistic features impact arousal or dominance. These findings show that transformers can successfully leverage linguistic information to improve their valence predictions, and that linguistic analysis should be included in their testing.

Johannes Wagner, Andreas Triantafyllopoulos, Hagen Wierstorf et al.

Recent advances in transformer-based architectures which are pre-trained in self-supervised manner have shown great promise in several machine learning tasks. In the audio domain, such architectures have also been successfully utilised in the field of speech emotion recognition (SER). However, existing works have not evaluated the influence of model size and pre-training data on downstream performance, and have shown limited attention to generalisation, robustness, fairness, and efficiency. The present contribution conducts a thorough analysis of these aspects on several pre-trained variants of wav2vec 2.0 and HuBERT that we fine-tuned on the dimensions arousal, dominance, and valence of MSP-Podcast, while additionally using IEMOCAP and MOSI to test cross-corpus generalisation. To the best of our knowledge, we obtain the top performance for valence prediction without use of explicit linguistic information, with a concordance correlation coefficient (CCC) of .638 on MSP-Podcast. Furthermore, our investigations reveal that transformer-based architectures are more robust to small perturbations compared to a CNN-based baseline and fair with respect to biological sex groups, but not towards individual speakers. Finally, we are the first to show that their extraordinary success on valence is based on implicit linguistic information learnt during fine-tuning of the transformer layers, which explains why they perform on-par with recent multimodal approaches that explicitly utilise textual information. Our findings collectively paint the following picture: transformer-based architectures constitute the new state-of-the-art in SER, but further advances are needed to mitigate remaining robustness and individual speaker issues. To make our findings reproducible, we release the best performing model to the community.

Oliver Schrüfer, Manuel Milling, Felix Burkhardt et al.

Uncertainty Quantification (UQ) is an important building block for the reliable use of neural networks in real-world scenarios, as it can be a useful tool in identifying faulty predictions. Speech emotion recognition (SER) models can suffer from particularly many sources of uncertainty, such as the ambiguity of emotions, Out-of-Distribution (OOD) data or, in general, poor recording conditions. Reliable UQ methods are thus of particular interest as in many SER applications no prediction is better than a faulty prediction. While the effects of label ambiguity on uncertainty are well documented in the literature, we focus our work on an evaluation of UQ methods for SER under common challenges in real-world application, such as corrupted signals, and the absence of speech. We show that simple UQ methods can already give an indication of the uncertainty of a prediction and that training with additional OOD data can greatly improve the identification of such signals.

Berkay Köprü, Mehrzad Mashal, Yigit Gurses et al.

Large language models (LLMs) have advanced natural language processing (NLP) skills such as through next-token prediction and self-attention, but their ability to integrate broad context also makes them prone to incorporating irrelevant information. Prior work has focused on semantic leakage, bias introduced by semantically irrelevant context. In this paper, we introduce expression leakage, a novel phenomenon where LLMs systematically generate sentimentally charged expressions that are semantically unrelated to the input context. To analyse the expression leakage, we collect a benchmark dataset along with a scheme to automatically generate a dataset from free-form text from common-crawl. In addition, we propose an automatic evaluation pipeline that correlates well with human judgment, which accelerates the benchmarking by decoupling from the need of annotation for each analysed model. Our experiments show that, as the model scales in the parameter space, the expression leakage reduces within the same LLM family. On the other hand, we demonstrate that expression leakage mitigation requires specific care during the model building process, and cannot be mitigated by prompting. In addition, our experiments indicate that, when negative sentiment is injected in the prompt, it disrupts the generation process more than the positive sentiment, causing a higher expression leakage rate.

Andreas Triantafyllopoulos, Uwe Reichel, Shuo Liu et al.

In this contribution, we investigate the effectiveness of deep fusion of text and audio features for categorical and dimensional speech emotion recognition (SER). We propose a novel, multistage fusion method where the two information streams are integrated in several layers of a deep neural network (DNN), and contrast it with a single-stage one where the streams are merged in a single point. Both methods depend on extracting summary linguistic embeddings from a pre-trained BERT model, and conditioning one or more intermediate representations of a convolutional model operating on log-Mel spectrograms. Experiments on the MSP-Podcast and IEMOCAP datasets demonstrate that the two fusion methods clearly outperform a shallow (late) fusion baseline and their unimodal constituents, both in terms of quantitative performance and qualitative behaviour. Overall, our multistage fusion shows better quantitative performance, surpassing alternatives on most of our evaluations. This illustrates the potential of multistage fusion in better assimilating text and audio information.

Anton Batliner, Stefan Steidl, Florian Eyben et al.

In this article, we study laughter found in child-robot interaction where it had not been prompted intentionally. Different types of laughter and speech-laugh are annotated and processed. In a descriptive part, we report on the position of laughter and speech-laugh in syntax and dialogue structure, and on communicative functions. In a second part, we report on automatic classification performance and on acoustic characteristics, based on extensive feature selection procedures.

Andreas Triantafyllopoulos, Hesam Sagha, Florian Eyben et al.

This paper describes audEERING's submissions as well as additional evaluations for the One-Minute-Gradual (OMG) emotion recognition challenge. We provide the results for audio and video processing on subject (in)dependent evaluations. On the provided Development set, we achieved 0.343 Concordance Correlation Coefficient (CCC) for arousal (from audio) and .401 for valence (from video).

Felix Weninger, Björn Schuller, Florian Eyben et al.

Transcription of broadcast news is an interesting and challenging application for large-vocabulary continuous speech recognition (LVCSR). We present in detail the structure of a manually segmented and annotated corpus including over 160 hours of German broadcast news, and propose it as an evaluation framework of LVCSR systems. We show our own experimental results on the corpus, achieved with a state-of-the-art LVCSR decoder, measuring the effect of different feature sets and decoding parameters, and thereby demonstrate that real-time decoding of our test set is feasible on a desktop PC at 9.2% word error rate.