Whisper is a powerful automatic speech recognition (ASR) model. Nevertheless, its zero-shot performance on low-resource speech requires further improvement. Child speech, as a representative type of low-resource speech, is leveraged for adaptation. Recently, parameter-efficient fine-tuning (PEFT) in NLP was shown to be comparable and even better than full fine-tuning, while only needing to tune a small set of trainable parameters. However, current PEFT methods have not been well examined for their effectiveness on Whisper. In this paper, only parameter composition types of PEFT approaches such as LoRA and Bitfit are investigated as they do not bring extra inference costs. Different popular PEFT methods are examined. Particularly, we compare LoRA and AdaLoRA and figure out the learnable rank coefficient is a good design. Inspired by the sparse rank distribution allocated by AdaLoRA, a novel PEFT approach Sparsely Shared LoRA (S2-LoRA) is proposed. The two low-rank decomposed matrices are globally shared. Each weight matrix only has to maintain its specific rank coefficients that are constrained to be sparse. Experiments on low-resource Chinese child speech show that with much fewer trainable parameters, S2-LoRA can achieve comparable in-domain adaptation performance to AdaLoRA and exhibit better generalization ability on out-of-domain data. In addition, the rank distribution automatically learned by S2-LoRA is found to have similar patterns to AdaLoRA's allocation.
Embodied agents, in the form of virtual agents or social robots, are rapidly becoming more widespread. In human-human interactions, humans use nonverbal behaviours to convey their attitudes, feelings, and intentions. Therefore, this capability is also required for embodied agents in order to enhance the quality and effectiveness of their interactions with humans. In this paper, we propose a novel framework that can generate sequences of joint angles from the speech text and speech audio utterances. Based on a conditional Generative Adversarial Network (GAN), our proposed neural network model learns the relationships between the co-speech gestures and both semantic and acoustic features from the speech input. In order to train our neural network model, we employ a public dataset containing co-speech gestures with corresponding speech audio utterances, which were captured from a single male native English speaker. The results from both objective and subjective evaluations demonstrate the efficacy of our gesture-generation framework for Robots and Embodied Agents.
The quality of automatic speech recognition (ASR) is typically measured by word error rate (WER). WER estimation is a task aiming to predict the WER of an ASR system, given a speech utterance and a transcription. This task has gained increasing attention while advanced ASR systems are trained on large amounts of data. In this case, WER estimation becomes necessary in many scenarios, for example, selecting training data with unknown transcription quality or estimating the testing performance of an ASR system without ground truth transcriptions. Facing large amounts of data, the computation efficiency of a WER estimator becomes essential in practical applications. However, previous works usually did not consider it as a priority. In this paper, a Fast WER estimator (Fe-WER) using self-supervised learning representation (SSLR) is introduced. The estimator is built upon SSLR aggregated by average pooling. The results show that Fe-WER outperformed the e-WER3 baseline relatively by 19.69% and 7.16% on Ted-Lium3 in both evaluation metrics of root mean square error and Pearson correlation coefficient, respectively. Moreover, the estimation weighted by duration was 10.43% when the target was 10.88%. Lastly, the inference speed was about 4x in terms of a real-time factor.
Modern ML systems ingest data aggregated from diverse sources, such as synthetic, human-annotated, and live customer traffic. Understanding \textit{which} examples are important to the performance of a learning algorithm is crucial for efficient model training. Recently, a growing body of literature has given rise to various "influence scores," which use training artifacts such as model confidence or checkpointed gradients to identify important subsets of data. However, these methods have primarily been developed in computer vision settings, and it remains unclear how well they generalize to language-based tasks using pretrained models. In this paper, we explore the applicability of influence scores in language classification tasks. We evaluate a diverse subset of these scores on the SNLI dataset by quantifying accuracy changes in response to pruning training data through random and influence-score-based sampling. We then stress-test one of the scores -- "variance of gradients" (VoG) from Agarwal et al. (2022) -- in an NLU model stack that was exposed to dynamic user speech patterns in a voice assistant type of setting. Our experiments demonstrate that in many cases, encoder-based language models can be finetuned on roughly 50% of the original data without degradation in performance metrics. Along the way, we summarize lessons learned from applying out-of-the-box implementations of influence scores, quantify the effects of noisy and class-imbalanced data, and offer recommendations on score-based sampling for better accuracy and training efficiency.
This paper explores the instruction fine-tuning technique for speech-to-semantic tasks by introducing a unified end-to-end (E2E) framework that generates target text conditioned on a task-related prompt for audio data. We pre-train the model using large and diverse data, where instruction-speech pairs are constructed via a text-to-speech (TTS) system. Extensive experiments demonstrate that our proposed model achieves state-of-the-art (SOTA) results on many benchmarks covering speech named entity recognition, speech sentiment analysis, speech question answering, and more, after fine-tuning. Furthermore, the proposed model achieves competitive performance in zero-shot and few-shot scenarios. To facilitate future work on instruction fine-tuning for speech-to-semantic tasks, we release our instruction dataset and code.
Languages usually switch within a multilingual speech signal, especially in a bilingual society. This phenomenon is referred to as code-switching (CS), making automatic speech recognition (ASR) challenging under a multilingual scenario. We propose to improve CS-ASR by biasing the hybrid CTC/attention ASR model with multi-level language information comprising frame- and token-level language posteriors. The interaction between various resolutions of language biases is subsequently explored in this work. We conducted experiments on datasets from the ASRU 2019 code-switching challenge. Compared to the baseline, the proposed interactive language biases (ILB) method achieves higher performance and ablation studies highlight the effects of different language biases and their interactions. In addition, the results presented indicate that language bias implicitly enhances internal language modeling, leading to performance degradation after employing an external language model.
Speech-driven 3D facial animation synthesis has been a challenging task both in industry and research. Recent methods mostly focus on deterministic deep learning methods meaning that given a speech input, the output is always the same. However, in reality, the non-verbal facial cues that reside throughout the face are non-deterministic in nature. In addition, majority of the approaches focus on 3D vertex based datasets and methods that are compatible with existing facial animation pipelines with rigged characters is scarce. To eliminate these issues, we present FaceDiffuser, a non-deterministic deep learning model to generate speech-driven facial animations that is trained with both 3D vertex and blendshape based datasets. Our method is based on the diffusion technique and uses the pre-trained large speech representation model HuBERT to encode the audio input. To the best of our knowledge, we are the first to employ the diffusion method for the task of speech-driven 3D facial animation synthesis. We have run extensive objective and subjective analyses and show that our approach achieves better or comparable results in comparison to the state-of-the-art methods. We also introduce a new in-house dataset that is based on a blendshape based rigged character. We recommend watching the accompanying supplementary video. The code and the dataset will be publicly available.
The generation of stylistic 3D facial animations driven by speech poses a significant challenge as it requires learning a many-to-many mapping between speech, style, and the corresponding natural facial motion. However, existing methods either employ a deterministic model for speech-to-motion mapping or encode the style using a one-hot encoding scheme. Notably, the one-hot encoding approach fails to capture the complexity of the style and thus limits generalization ability. In this paper, we propose DiffPoseTalk, a generative framework based on the diffusion model combined with a style encoder that extracts style embeddings from short reference videos. During inference, we employ classifier-free guidance to guide the generation process based on the speech and style. We extend this to include the generation of head poses, thereby enhancing user perception. Additionally, we address the shortage of scanned 3D talking face data by training our model on reconstructed 3DMM parameters from a high-quality, in-the-wild audio-visual dataset. Our extensive experiments and user study demonstrate that our approach outperforms state-of-the-art methods. The code and dataset will be made publicly available.
Self-supervised pre-trained speech models have strongly improved speech recognition, yet they are still sensitive to domain shifts and accented or atypical speech. Many of these models rely on quantisation or clustering to learn discrete acoustic units. We propose to correct the discovered discrete units for accented speech back to a standard pronunciation in an unsupervised manner. A masked language model is trained on discrete units from a standard accent and iteratively corrects an accented token sequence by masking unexpected cluster sequences and predicting their common variant. Small accent adapter blocks are inserted in the pre-trained model and fine-tuned by predicting the corrected clusters, which leads to an increased robustness of the pre-trained model towards a target accent, and this without supervision. We are able to improve a state-of-the-art HuBERT Large model on a downstream accented speech recognition task by altering the training regime with the proposed method.
Generative Adversarial Network (GAN) based vocoders are superior in inference speed and synthesis quality when reconstructing an audible waveform from an acoustic representation. This study focuses on improving the discriminator to promote GAN-based vocoders. Most existing time-frequency-representation-based discriminators are rooted in Short-Time Fourier Transform (STFT), whose time-frequency resolution in a spectrogram is fixed, making it incompatible with signals like singing voices that require flexible attention for different frequency bands. Motivated by that, our study utilizes the Constant-Q Transform (CQT), which owns dynamic resolution among frequencies, contributing to a better modeling ability in pitch accuracy and harmonic tracking. Specifically, we propose a Multi-Scale Sub-Band CQT (MS-SB-CQT) Discriminator, which operates on the CQT spectrogram at multiple scales and performs sub-band processing according to different octaves. Experiments conducted on both speech and singing voices confirm the effectiveness of our proposed method. Moreover, we also verified that the CQT-based and the STFT-based discriminators could be complementary under joint training. Specifically, enhanced by the proposed MS-SB-CQT and the existing MS-STFT Discriminators, the MOS of HiFi-GAN can be boosted from 3.27 to 3.87 for seen singers and from 3.40 to 3.78 for unseen singers.