Recent advances in eXplainable AI (XAI) have provided new insights into how models for vision, language, and tabular data operate. However, few approaches exist for understanding speech models. Existing work focuses on a few spoken language understanding (SLU) tasks, and explanations are difficult to interpret for most users. We introduce a new approach to explain speech classification models. We generate easy-to-interpret explanations via input perturbation on two information levels. 1) Word-level explanations reveal how each word-related audio segment impacts the outcome. 2) Paralinguistic features (e.g., prosody and background noise) answer the counterfactual: ``What would the model prediction be if we edited the audio signal in this way?'' We validate our approach by explaining two state-of-the-art SLU models on two speech classification tasks in English and Italian. Our findings demonstrate that the explanations are faithful to the model's inner workings and plausible to humans. Our method and findings pave the way for future research on interpreting speech models.
In this paper, we propose methods to build a powerful and efficient Image-to-Speech captioning (Im2Sp) model. To this end, we start with importing the rich knowledge related to image comprehension and language modeling from a large-scale pre-trained vision-language model into Im2Sp. We set the output of the proposed Im2Sp as discretized speech units, i.e., the quantized speech features of a self-supervised speech model. The speech units mainly contain linguistic information while suppressing other characteristics of speech. This allows us to incorporate the language modeling capability of the pre-trained vision-language model into the spoken language modeling of Im2Sp. With the vision-language pre-training strategy, we set new state-of-the-art Im2Sp performances on two widely used benchmark databases, COCO and Flickr8k. Then, we further improve the efficiency of the Im2Sp model. Similar to the speech unit case, we convert the original image into image units, which are derived through vector quantization of the raw image. With these image units, we can drastically reduce the required data storage for saving image data to just 0.8% when compared to the original image data in terms of bits. Demo page: https://ms-dot-k.github.io/Image-to-Speech-Captioning.
Speech emotion recognition has evolved from research to practical applications. Previous studies of emotion recognition from speech have focused on developing models on certain datasets like IEMOCAP. The lack of data in the domain of emotion modeling emerges as a challenge to evaluate models in the other dataset, as well as to evaluate speech emotion recognition models that work in a multilingual setting. This paper proposes an ensemble learning to fuse results of pre-trained models for emotion share recognition from speech. The models were chosen to accommodate multilingual data from English and Spanish. The results show that ensemble learning can improve the performance of the baseline model with a single model and the previous best model from the late fusion. The performance is measured using the Spearman rank correlation coefficient since the task is a regression problem with ranking values. A Spearman rank correlation coefficient of 0.537 is reported for the test set, while for the development set, the score is 0.524. These scores are higher than the previous study of a fusion method from monolingual data, which achieved scores of 0.476 for the test and 0.470 for the development.
Recent studies have increasingly acknowledged the advantages of incorporating visual data into speech enhancement (SE) systems. In this paper, we introduce a novel audio-visual SE approach, termed DCUC-Net (deep complex U-Net with conformer network). The proposed DCUC-Net leverages complex domain features and a stack of conformer blocks. The encoder and decoder of DCUC-Net are designed using a complex U-Net-based framework. The audio and visual signals are processed using a complex encoder and a ResNet-18 model, respectively. These processed signals are then fused using the conformer blocks and transformed into enhanced speech waveforms via a complex decoder. The conformer blocks consist of a combination of self-attention mechanisms and convolutional operations, enabling DCUC-Net to effectively capture both global and local audio-visual dependencies. Our experimental results demonstrate the effectiveness of DCUC-Net, as it outperforms the baseline model from the COG-MHEAR AVSE Challenge 2023 by a notable margin of 0.14 in terms of PESQ. Additionally, the proposed DCUC-Net performs comparably to a state-of-the-art model and outperforms all other compared models on the Taiwan Mandarin speech with video (TMSV) dataset.
The emergence of large language models (LLMs) has sparked significant interest in extending their remarkable language capabilities to speech. However, modality alignment between speech and text still remains an open problem. Current solutions can be categorized into two strategies. One is a cascaded approach where outputs (tokens or states) of a separately trained speech recognition system are used as inputs for LLMs, which limits their potential in modeling alignment between speech and text. The other is an end-to-end approach that relies on speech instruction data, which is very difficult to collect in large quantities. In this paper, we address these issues and propose the BLSP approach that Bootstraps Language-Speech Pre-training via behavior alignment of continuation writing. We achieve this by learning a lightweight modality adapter between a frozen speech encoder and an LLM, ensuring that the LLM exhibits the same generation behavior regardless of the modality of input: a speech segment or its transcript. The training process can be divided into two steps. The first step prompts an LLM to generate texts with speech transcripts as prefixes, obtaining text continuations. In the second step, these continuations are used as supervised signals to train the modality adapter in an end-to-end manner. We demonstrate that this straightforward process can extend the capabilities of LLMs to speech, enabling speech recognition, speech translation, spoken language understanding, and speech conversation, even in zero-shot cross-lingual scenarios.
Transformer-based models excel in speech recognition. Existing efforts to optimize Transformer inference, typically for long-context applications, center on simplifying attention score calculations. However, streaming speech recognition models usually process a limited number of tokens each time, making attention score calculation less of a bottleneck. Instead, the bottleneck lies in the linear projection layers of multi-head attention and feedforward networks, constituting a substantial portion of the model size and contributing significantly to computation, memory, and power usage. To address this bottleneck, we propose folding attention, a technique targeting these linear layers, significantly reducing model size and improving memory and power efficiency. Experiments on on-device Transformer-based streaming speech recognition models show that folding attention reduces model size (and corresponding memory consumption) by up to 24% and power consumption by up to 23%, all without compromising model accuracy or computation overhead.
Blockwise self-attentional encoder models have recently emerged as one promising end-to-end approach to simultaneous speech translation. These models employ a blockwise beam search with hypothesis reliability scoring to determine when to wait for more input speech before translating further. However, this method maintains multiple hypotheses until the entire speech input is consumed -- this scheme cannot directly show a single \textit{incremental} translation to users. Further, this method lacks mechanisms for \textit{controlling} the quality vs. latency tradeoff. We propose a modified incremental blockwise beam search incorporating local agreement or hold-$n$ policies for quality-latency control. We apply our framework to models trained for online or offline translation and demonstrate that both types can be effectively used in online mode. Experimental results on MuST-C show 0.6-3.6 BLEU improvement without changing latency or 0.8-1.4 s latency improvement without changing quality.
Speech enhancement seeks to extract clean speech from noisy signals. Traditional deep learning methods face two challenges: efficiently using information in long speech sequences and high computational costs. To address these, we introduce the Spiking Structured State Space Model (Spiking-S4). This approach merges the energy efficiency of Spiking Neural Networks (SNN) with the long-range sequence modeling capabilities of Structured State Space Models (S4), offering a compelling solution. Evaluation on the DNS Challenge and VoiceBank+Demand Datasets confirms that Spiking-S4 rivals existing Artificial Neural Network (ANN) methods but with fewer computational resources, as evidenced by reduced parameters and Floating Point Operations (FLOPs).
Recent advancements in Automatic Speech Recognition (ASR) systems, exemplified by Whisper, have demonstrated the potential of these systems to approach human-level performance given sufficient data. However, this progress doesn't readily extend to ASR for children due to the limited availability of suitable child-specific databases and the distinct characteristics of children's speech. A recent study investigated leveraging the My Science Tutor (MyST) children's speech corpus to enhance Whisper's performance in recognizing children's speech. This paper builds on these findings by enhancing the utility of the MyST dataset through more efficient data preprocessing. We also highlight important challenges towards improving children's ASR performance. The results showcase the viable and efficient integration of Whisper for effective children's speech recognition.
The application of self-supervision to speech representation learning has garnered significant interest in recent years, due to its scalability to large amounts of unlabeled data. However, much progress, both in terms of pre-training and downstream evaluation, has remained concentrated in monolingual models that only consider English. Few models consider other languages, and even fewer consider indigenous ones. In our submission to the New Language Track of the ASRU 2023 ML-SUPERB Challenge, we present an ASR corpus for Quechua, an indigenous South American Language. We benchmark the efficacy of large SSL models on Quechua, along with 6 other indigenous languages such as Guarani and Bribri, on low-resource ASR. Our results show surprisingly strong performance by state-of-the-art SSL models, showing the potential generalizability of large-scale models to real-world data.