Good practices for evaluation of machine learning systems

Many development decisions affect the results obtained from ML experiments: training data, features, model architecture, hyperparameters, test data, etc. Among these aspects, arguably the most important design decisions are those that involve the evaluation procedure. This procedure is what determines whether the conclusions drawn from the experiments will or will not generalize to unseen data and whether they will be relevant to the application of interest. If the data is incorrectly selected, the wrong metric is chosen for evaluation or the significance of the comparisons between models is overestimated, conclusions may be misleading or result in suboptimal development decisions. To avoid such problems, the evaluation protocol should be very carefully designed before experimentation starts. In this work we discuss the main aspects involved in the design of the evaluation protocol: data selection, metric selection, and statistical significance. This document is not meant to be an exhaustive tutorial on each of these aspects. Instead, the goal is to explain the main guidelines that should be followed in each case. We include examples taken from the speech processing field, and provide a list of common mistakes related to each aspect.

Self-supervised Speech Representations Still Struggle with African American Vernacular English

Underperformance of ASR systems for speakers of African American Vernacular English (AAVE) and other marginalized language varieties is a well-documented phenomenon, and one that reinforces the stigmatization of these varieties. We investigate whether or not the recent wave of Self-Supervised Learning (SSL) speech models can close the gap in ASR performance between AAVE and Mainstream American English (MAE). We evaluate four SSL models (wav2vec 2.0, HuBERT, WavLM, and XLS-R) on zero-shot Automatic Speech Recognition (ASR) for these two varieties and find that these models perpetuate the bias in performance against AAVE. Additionally, the models have higher word error rates on utterances with more phonological and morphosyntactic features of AAVE. Despite the success of SSL speech models in improving ASR for low resource varieties, SSL pre-training alone may not bridge the gap between AAVE and MAE. Our code is publicly available at https://github.com/cmu-llab/s3m-aave.

As Biased as You Measure: Methodological Pitfalls of Bias Evaluations in Speaker Verification Research

Detecting and mitigating bias in speaker verification systems is important, as datasets, processing choices and algorithms can lead to performance differences that systematically favour some groups of people while disadvantaging others. Prior studies have thus measured performance differences across groups to evaluate bias. However, when comparing results across studies, it becomes apparent that they draw contradictory conclusions, hindering progress in this area. In this paper we investigate how measurement impacts the outcomes of bias evaluations. We show empirically that bias evaluations are strongly influenced by base metrics that measure performance, by the choice of ratio or difference-based bias measure, and by the aggregation of bias measures into meta-measures. Based on our findings, we recommend the use of ratio-based bias measures, in particular when the values of base metrics are small, or when base metrics with different orders of magnitude need to be compared.

Improving child speech recognition with augmented child-like speech

State-of-the-art ASRs show suboptimal performance for child speech. The scarcity of child speech limits the development of child speech recognition (CSR). Therefore, we studied child-to-child voice conversion (VC) from existing child speakers in the dataset and additional (new) child speakers via monolingual and cross-lingual (Dutch-to-German) VC, respectively. The results showed that cross-lingual child-to-child VC significantly improved child ASR performance. Experiments on the impact of the quantity of child-to-child cross-lingual VC-generated data on fine-tuning (FT) ASR models gave the best results with two-fold augmentation for our FT-Conformer model and FT-Whisper model which reduced WERs with ~3% absolute compared to the baseline, and with six-fold augmentation for the model trained from scratch, which improved by an absolute 3.6% WER. Moreover, using a small amount of "high-quality" VC-generated data achieved similar results to those of our best-FT models.

Exploring data augmentation in bias mitigation against non-native-accented speech

Automatic speech recognition (ASR) should serve every speaker, not only the majority ``standard'' speakers of a language. In order to build inclusive ASR, mitigating the bias against speaker groups who speak in a ``non-standard'' or ``diverse'' way is crucial. We aim to mitigate the bias against non-native-accented Flemish in a Flemish ASR system. Since this is a low-resource problem, we investigate the optimal type of data augmentation, i.e., speed/pitch perturbation, cross-lingual voice conversion-based methods, and SpecAugment, applied to both native Flemish and non-native-accented Flemish, for bias mitigation. The results showed that specific types of data augmentation applied to both native and non-native-accented speech improve non-native-accented ASR while applying data augmentation to the non-native-accented speech is more conducive to bias reduction. Combining both gave the largest bias reduction for human-machine interaction (HMI) as well as read-type speech.

Improving Whispered Speech Recognition Performance using Pseudo-whispered based Data Augmentation

Whispering is a distinct form of speech known for its soft, breathy, and hushed characteristics, often used for private communication. The acoustic characteristics of whispered speech differ substantially from normally phonated speech and the scarcity of adequate training data leads to low automatic speech recognition (ASR) performance. To address the data scarcity issue, we use a signal processing-based technique that transforms the spectral characteristics of normal speech to those of pseudo-whispered speech. We augment an End-to-End ASR with pseudo-whispered speech and achieve an 18.2% relative reduction in word error rate for whispered speech compared to the baseline. Results for the individual speaker groups in the wTIMIT database show the best results for US English. Further investigation showed that the lack of glottal information in whispered speech has the largest impact on whispered speech ASR performance.

The Multimodal Information Based Speech Processing (MISP) 2023 Challenge: Audio-Visual Target Speaker Extraction

Previous Multimodal Information based Speech Processing (MISP) challenges mainly focused on audio-visual speech recognition (AVSR) with commendable success. However, the most advanced back-end recognition systems often hit performance limits due to the complex acoustic environments. This has prompted a shift in focus towards the Audio-Visual Target Speaker Extraction (AVTSE) task for the MISP 2023 challenge in ICASSP 2024 Signal Processing Grand Challenges. Unlike existing audio-visual speech enhance-ment challenges primarily focused on simulation data, the MISP 2023 challenge uniquely explores how front-end speech processing, combined with visual clues, impacts back-end tasks in real-world scenarios. This pioneering effort aims to set the first benchmark for the AVTSE task, offering fresh insights into enhancing the ac-curacy of back-end speech recognition systems through AVTSE in challenging and real acoustic environments. This paper delivers a thorough overview of the task setting, dataset, and baseline system of the MISP 2023 challenge. It also includes an in-depth analysis of the challenges participants may encounter. The experimental results highlight the demanding nature of this task, and we look forward to the innovative solutions participants will bring forward.

Using Data Augmentations and VTLN to Reduce Bias in Dutch End-to-End Speech Recognition Systems

Speech technology has improved greatly for norm speakers, i.e., adult native speakers of a language without speech impediments or strong accents. However, non-norm or diverse speaker groups show a distinct performance gap with norm speakers, which we refer to as bias. In this work, we aim to reduce bias against different age groups and non-native speakers of Dutch. For an end-to-end (E2E) ASR system, we use state-of-the-art speed perturbation and spectral augmentation as data augmentation techniques and explore Vocal Tract Length Normalization (VTLN) to normalise for spectral differences due to differences in anatomy. The combination of data augmentation and VTLN reduced the average WER and bias across various diverse speaker groups by 6.9% and 3.9%, respectively. The VTLN model trained on Dutch was also effective in improving performance of Mandarin Chinese child speech, thus, showing generalisability across languages

Predicting within and across language phoneme recognition performance of self-supervised learning speech pre-trained models

In this work, we analyzed and compared speech representations extracted from different frozen self-supervised learning (SSL) speech pre-trained models on their ability to capture articulatory features (AF) information and their subsequent prediction of phone recognition performance for within and across language scenarios. Specifically, we compared CPC, wav2vec 2.0, and HuBert. First, frame-level AF probing tasks were implemented. Subsequently, phone-level end-to-end ASR systems for phoneme recognition tasks were implemented, and the performance on the frame-level AF probing task and the phone accuracy were correlated. Compared to the conventional speech representation MFCC, all SSL pre-trained speech representations captured more AF information, and achieved better phoneme recognition performance within and across languages, with HuBert performing best. The frame-level AF probing task is a good predictor of phoneme recognition performance, showing the importance of capturing AF information in the speech representations. Compared with MFCC, in the within-language scenario, the performance of these SSL speech pre-trained models on AF probing tasks achieved a maximum relative increase of 34.4%, and it resulted in the lowest PER of 10.2%. In the cross-language scenario, the maximum relative increase of 26.7% also resulted in the lowest PER of 23.0%.

Manipulation of oral cancer speech using neural articulatory synthesis

We present an articulatory synthesis framework for the synthesis and manipulation of oral cancer speech for clinical decision making and alleviation of patient stress. Objective and subjective evaluations demonstrate that the framework has acceptable naturalness and is worth further investigation. A subsequent subjective vowel and consonant identification experiment showed that the articulatory synthesis system can manipulate the articulatory trajectories so that the synthesised speech reproduces problems present in the ground truth oral cancer speech.