Sefik Emre Eskimez

Tuochao Chen, Qirui Wang, Bohan Wu et al.

Extracting the speech of participants in a conversation amidst interfering speakers and noise presents a challenging problem. In this paper, we introduce the novel task of target conversation extraction, where the goal is to extract the audio of a target conversation based on the speaker embedding of one of its participants. To accomplish this, we propose leveraging temporal patterns inherent in human conversations, particularly turn-taking dynamics, which uniquely characterize speakers engaged in conversation and distinguish them from interfering speakers and noise. Using neural networks, we show the feasibility of our approach on English and Mandarin conversation datasets. In the presence of interfering speakers, our results show an 8.19 dB improvement in signal-to-noise ratio for 2-speaker conversations and a 7.92 dB improvement for 2-4-speaker conversations. Code, dataset available at https://github.com/chentuochao/Target-Conversation-Extraction.

Xiaofei Wang, Manthan Thakker, Zhuo Chen et al.

Recent advancements in generative speech models based on audio-text prompts have enabled remarkable innovations like high-quality zero-shot text-to-speech. However, existing models still face limitations in handling diverse audio-text speech generation tasks involving transforming input speech and processing audio captured in adverse acoustic conditions. This paper introduces SpeechX, a versatile speech generation model capable of zero-shot TTS and various speech transformation tasks, dealing with both clean and noisy signals. SpeechX combines neural codec language modeling with multi-task learning using task-dependent prompting, enabling unified and extensible modeling and providing a consistent way for leveraging textual input in speech enhancement and transformation tasks. Experimental results show SpeechX's efficacy in various tasks, including zero-shot TTS, noise suppression, target speaker extraction, speech removal, and speech editing with or without background noise, achieving comparable or superior performance to specialized models across tasks. See https://aka.ms/speechx for demo samples.

Zhuo Chen, Naoyuki Kanda, Jian Wu et al.

Self-supervised learning (SSL) methods such as WavLM have shown promising speech separation (SS) results in small-scale simulation-based experiments. In this work, we extend the exploration of the SSL-based SS by massively scaling up both the pre-training data (more than 300K hours) and fine-tuning data (10K hours). We also investigate various techniques to efficiently integrate the pre-trained model with the SS network under a limited computation budget, including a low frame rate SSL model training setup and a fine-tuning scheme using only the part of the pre-trained model. Compared with a supervised baseline and the WavLM-based SS model using feature embeddings obtained with the previously released 94K hours trained WavLM, our proposed model obtains 15.9% and 11.2% of relative word error rate (WER) reductions, respectively, for a simulated far-field speech mixture test set. For conversation transcription on real meeting recordings using continuous speech separation, the proposed model achieves 6.8% and 10.6% of relative WER reductions over the purely supervised baseline on AMI and ICSI evaluation sets, respectively, while reducing the computational cost by 38%.

Haibin Wu, Xiaofei Wang, Sefik Emre Eskimez et al.

People change their tones of voice, often accompanied by nonverbal vocalizations (NVs) such as laughter and cries, to convey rich emotions. However, most text-to-speech (TTS) systems lack the capability to generate speech with rich emotions, including NVs. This paper introduces EmoCtrl-TTS, an emotion-controllable zero-shot TTS that can generate highly emotional speech with NVs for any speaker. EmoCtrl-TTS leverages arousal and valence values, as well as laughter embeddings, to condition the flow-matching-based zero-shot TTS. To achieve high-quality emotional speech generation, EmoCtrl-TTS is trained using more than 27,000 hours of expressive data curated based on pseudo-labeling. Comprehensive evaluations demonstrate that EmoCtrl-TTS excels in mimicking the emotions of audio prompts in speech-to-speech translation scenarios. We also show that EmoCtrl-TTS can capture emotion changes, express strong emotions, and generate various NVs in zero-shot TTS. See https://aka.ms/emoctrl-tts for demo samples.

Xiaofei Wang, Dongmei Wang, Naoyuki Kanda et al.

Existing multi-channel continuous speech separation (CSS) models are heavily dependent on supervised data - either simulated data which causes data mismatch between the training and real-data testing, or the real transcribed overlapping data, which is difficult to be acquired, hindering further improvements in the conversational/meeting transcription tasks. In this paper, we propose a three-stage training scheme for the CSS model that can leverage both supervised data and extra large-scale unsupervised real-world conversational data. The scheme consists of two conventional training approaches -- pre-training using simulated data and ASR-loss-based training using transcribed data -- and a novel continuous semi-supervised training between the two, in which the CSS model is further trained by using real data based on the teacher-student learning framework. We apply this scheme to an array-geometry-agnostic CSS model, which can use the multi-channel data collected from any microphone array. Large-scale meeting transcription experiments are carried out on both Microsoft internal meeting data and the AMI meeting corpus. The steady improvement by each training stage has been observed, showing the effect of the proposed method that enables leveraging real conversational data for CSS model training.

Vidya Srinivas, Malek Itani, Tuochao Chen et al.

Models for low-latency, streaming applications could benefit from the knowledge capacity of larger models, but edge devices cannot run these models due to resource constraints. A possible solution is to transfer hints during inference from a large model running remotely to a small model running on-device. However, this incurs a communication delay that breaks real-time requirements and does not guarantee that both models will operate on the same data at the same time. We propose knowledge boosting, a novel technique that allows a large model to operate on time-delayed input during inference, while still boosting small model performance. Using a streaming neural network that processes 8 ms chunks, we evaluate different speech separation and enhancement tasks with communication delays of up to six chunks or 48 ms. Our results show larger gains where the performance gap between the small and large models is wide, demonstrating a promising method for large-small model collaboration for low-latency applications. Code, dataset, and audio samples available at https://knowledgeboosting.cs.washington.edu/.

Harishchandra Dubey, Vishak Gopal, Ross Cutler et al.

The Deep Noise Suppression (DNS) challenge is designed to foster innovation in the area of noise suppression to achieve superior perceptual speech quality. This is the 4th DNS challenge, with the previous editions held at INTERSPEECH 2020, ICASSP 2021, and INTERSPEECH 2021. We open-source datasets and test sets for researchers to train their deep noise suppression models, as well as a subjective evaluation framework based on ITU-T P.835 to rate and rank-order the challenge entries. We provide access to DNSMOS P.835 and word accuracy (WAcc) APIs to challenge participants to help with iterative model improvements. In this challenge, we introduced the following changes: (i) Included mobile device scenarios in the blind test set; (ii) Included a personalized noise suppression track with baseline; (iii) Added WAcc as an objective metric; (iv) Included DNSMOS P.835; (v) Made the training datasets and test sets fullband (48 kHz). We use an average of WAcc and subjective scores P.835 SIG, BAK, and OVRL to get the final score for ranking the DNS models. We believe that as a research community, we still have a long way to go in achieving excellent speech quality in challenging noisy real-world scenarios.

Junwei Liao, Sefik Emre Eskimez, Liyang Lu et al.

Modern Automatic Speech Recognition (ASR) systems can achieve high performance in terms of recognition accuracy. However, a perfectly accurate transcript still can be challenging to read due to grammatical errors, disfluency, and other errata common in spoken communication. Many downstream tasks and human readers rely on the output of the ASR system; therefore, errors introduced by the speaker and ASR system alike will be propagated to the next task in the pipeline. In this work, we propose a novel NLP task called ASR post-processing for readability (APR) that aims to transform the noisy ASR output into a readable text for humans and downstream tasks while maintaining the semantic meaning of the speaker. In addition, we describe a method to address the lack of task-specific data by synthesizing examples for the APR task using the datasets collected for Grammatical Error Correction (GEC) followed by text-to-speech (TTS) and ASR. Furthermore, we propose metrics borrowed from similar tasks to evaluate performance on the APR task. We compare fine-tuned models based on several open-sourced and adapted pre-trained models with the traditional pipeline method. Our results suggest that finetuned models improve the performance on the APR task significantly, hinting at the potential benefits of using APR systems. We hope that the read, understand, and rewrite approach of our work can serve as a basis that many NLP tasks and human readers can benefit from.

Sefik Emre Eskimez, Kenneth Imade, Na Yang et al. · apple-ml

The fact that emotions play a vital role in social interactions, along with the demand for novel human-computer interaction applications, have led to the development of a number of automatic emotion classification systems. However, it is still debatable whether the performance of such systems can compare with human coders. To address this issue, in this study, we present a comprehensive comparison in a speech-based emotion classification task between 138 Amazon Mechanical Turk workers (Turkers) and a state-of-the-art automatic computer system. The comparison includes classifying speech utterances into six emotions (happy, neutral, sad, anger, disgust and fear), into three arousal classes (active, passive, and neutral), and into three valence classes (positive, negative, and neutral). The results show that the computer system outperforms the naive Turkers in almost all cases. Furthermore, the computer system can increase the classification accuracy by rejecting to classify utterances for which it is not confident, while the Turkers do not show a significantly higher classification accuracy on their confident utterances versus unconfident ones.

Naoyuki Kanda, Xiaofei Wang, Sefik Emre Eskimez et al.

Laughter is one of the most expressive and natural aspects of human speech, conveying emotions, social cues, and humor. However, most text-to-speech (TTS) systems lack the ability to produce realistic and appropriate laughter sounds, limiting their applications and user experience. While there have been prior works to generate natural laughter, they fell short in terms of controlling the timing and variety of the laughter to be generated. In this work, we propose ELaTE, a zero-shot TTS that can generate natural laughing speech of any speaker based on a short audio prompt with precise control of laughter timing and expression. Specifically, ELaTE works on the audio prompt to mimic the voice characteristic, the text prompt to indicate the contents of the generated speech, and the input to control the laughter expression, which can be either the start and end times of laughter, or the additional audio prompt that contains laughter to be mimicked. We develop our model based on the foundation of conditional flow-matching-based zero-shot TTS, and fine-tune it with frame-level representation from a laughter detector as additional conditioning. With a simple scheme to mix small-scale laughter-conditioned data with large-scale pre-training data, we demonstrate that a pre-trained zero-shot TTS model can be readily fine-tuned to generate natural laughter with precise controllability, without losing any quality of the pre-trained zero-shot TTS model. Through objective and subjective evaluations, we show that ELaTE can generate laughing speech with significantly higher quality and controllability compared to conventional models. See https://aka.ms/elate/ for demo samples.

Malek Itani, Ashton Graves, Sefik Emre Eskimez et al.

We present the first neural target speech extraction (TSE) system that uses human feedback for iterative refinement. Our approach allows users to mark specific segments of the TSE output, generating an edit mask. The refinement system then improves the marked sections while preserving unmarked regions. Since large-scale datasets of human-marked errors are difficult to collect, we generate synthetic datasets using various automated masking functions and train models on each. Evaluations show that models trained with noise power-based masking (in dBFS) and probabilistic thresholding perform best, aligning with human annotations. In a study with 22 participants, users showed a preference for refined outputs over baseline TSE. Our findings demonstrate that human-in-the-loop refinement is a promising approach for improving the performance of neural speech extraction.

Xiaofei Wang, Sefik Emre Eskimez, Manthan Thakker et al.

Recently, zero-shot text-to-speech (TTS) systems, capable of synthesizing any speaker's voice from a short audio prompt, have made rapid advancements. However, the quality of the generated speech significantly deteriorates when the audio prompt contains noise, and limited research has been conducted to address this issue. In this paper, we explored various strategies to enhance the quality of audio generated from noisy audio prompts within the context of flow-matching-based zero-shot TTS. Our investigation includes comprehensive training strategies: unsupervised pre-training with masked speech denoising, multi-speaker detection and DNSMOS-based data filtering on the pre-training data, and fine-tuning with random noise mixing. The results of our experiments demonstrate significant improvements in intelligibility, speaker similarity, and overall audio quality compared to the approach of applying speech enhancement to the audio prompt.

Kenichi Kumatani, Dimitrios Dimitriadis, Yashesh Gaur et al.

In this work, we develop new self-learning techniques with an attention-based sequence-to-sequence (seq2seq) model for automatic speech recognition (ASR). For untranscribed speech data, the hypothesis from an ASR system must be used as a label. However, the imperfect ASR result makes unsupervised learning difficult to consistently improve recognition performance especially in the case that multiple powerful teacher models are unavailable. In contrast to conventional unsupervised learning approaches, we adopt the \emph{multi-task learning} (MTL) framework where the $n$-th best ASR hypothesis is used as the label of each task. The seq2seq network is updated through the MTL framework so as to find the common representation that can cover multiple hypotheses. By doing so, the effect of the \emph{hard-decision} errors can be alleviated. We first demonstrate the effectiveness of our self-learning methods through ASR experiments in an accent adaptation task between the US and British English speech. Our experiment results show that our method can reduce the WER on the British speech data from 14.55\% to 10.36\% compared to the baseline model trained with the US English data only. Moreover, we investigate the effect of our proposed methods in a federated learning scenario.

Wangyou Zhang, Zhuo Chen, Naoyuki Kanda et al.

Multi-talker conversational speech processing has drawn many interests for various applications such as meeting transcription. Speech separation is often required to handle overlapped speech that is commonly observed in conversation. Although the original utterancelevel permutation invariant training-based continuous speech separation approach has proven to be effective in various conditions, it lacks the ability to leverage the long-span relationship of utterances and is computationally inefficient due to the highly overlapped sliding windows. To overcome these drawbacks, we propose a novel training scheme named Group-PIT, which allows direct training of the speech separation models on the long-form speech with a low computational cost for label assignment. Two different speech separation approaches with Group-PIT are explored, including direct long-span speech separation and short-span speech separation with long-span tracking. The experiments on the simulated meeting-style data demonstrate the effectiveness of our proposed approaches, especially in dealing with a very long speech input.

Hassan Taherian, Sefik Emre Eskimez, Takuya Yoshioka et al.

With the recent surge of video conferencing tools usage, providing high-quality speech signals and accurate captions have become essential to conduct day-to-day business or connect with friends and families. Single-channel personalized speech enhancement (PSE) methods show promising results compared with the unconditional speech enhancement (SE) methods in these scenarios due to their ability to remove interfering speech in addition to the environmental noise. In this work, we leverage spatial information afforded by microphone arrays to improve such systems' performance further. We investigate the relative importance of speaker embeddings and spatial features. Moreover, we propose a new causal array-geometry-agnostic multi-channel PSE model, which can generate a high-quality enhanced signal from arbitrary microphone geometry. Experimental results show that the proposed geometry agnostic model outperforms the model trained on a specific microphone array geometry in both speech quality and automatic speech recognition accuracy. We also demonstrate the effectiveness of the proposed approach for unseen array geometries.

Sefik Emre Eskimez, Takuya Yoshioka, Huaming Wang et al.

Personalized speech enhancement (PSE) models utilize additional cues, such as speaker embeddings like d-vectors, to remove background noise and interfering speech in real-time and thus improve the speech quality of online video conferencing systems for various acoustic scenarios. In this work, we propose two neural networks for PSE that achieve superior performance to the previously proposed VoiceFilter. In addition, we create test sets that capture a variety of scenarios that users can encounter during video conferencing. Furthermore, we propose a new metric to measure the target speaker over-suppression (TSOS) problem, which was not sufficiently investigated before despite its critical importance in deployment. Besides, we propose multi-task training with a speech recognition back-end. Our results show that the proposed models can yield better speech recognition accuracy, speech intelligibility, and perceptual quality than the baseline models, and the multi-task training can alleviate the TSOS issue in addition to improving the speech recognition accuracy.

Zhuohuang Zhang, Takuya Yoshioka, Naoyuki Kanda et al.

Continuous speech separation (CSS) aims to separate overlapping voices from a continuous influx of conversational audio containing an unknown number of utterances spoken by an unknown number of speakers. A common application scenario is transcribing a meeting conversation recorded by a microphone array. Prior studies explored various deep learning models for time-frequency mask estimation, followed by a minimum variance distortionless response (MVDR) filter to improve the automatic speech recognition (ASR) accuracy. The performance of these methods is fundamentally upper-bounded by MVDR's spatial selectivity. Recently, the all deep learning MVDR (ADL-MVDR) model was proposed for neural beamforming and demonstrated superior performance in a target speech extraction task using pre-segmented input. In this paper, we further adapt ADL-MVDR to the CSS task with several enhancements to enable end-to-end neural beamforming. The proposed system achieves significant word error rate reduction over a baseline spectral masking system on the LibriCSS dataset. Moreover, the proposed neural beamformer is shown to be comparable to a state-of-the-art MVDR-based system in real meeting transcription tasks, including AMI, while showing potentials to further simplify the runtime implementation and reduce the system latency with frame-wise processing.

Dimitrios Dimitriadis, Kenichi Kumatani, Robert Gmyr et al.

In this paper, a new learning algorithm for Federated Learning (FL) is introduced. The proposed scheme is based on a weighted gradient aggregation using two-step optimization to offer a flexible training pipeline. Herein, two different flavors of the aggregation method are presented, leading to an order of magnitude improvement in convergence speed compared to other distributed or FL training algorithms like BMUF and FedAvg. Further, the aggregation algorithm acts as a regularizer of the gradient quality. We investigate the effect of our FL algorithm in supervised and unsupervised Speech Recognition (SR) scenarios. The experimental validation is performed based on three tasks: first, the LibriSpeech task showing a speed-up of 7x and 6% word error rate reduction (WERR) compared to the baseline results. The second task is based on session adaptation providing 20% WERR over a powerful LAS model. Finally, our unsupervised pipeline is applied to the conversational SR task. The proposed FL system outperforms the baseline systems in both convergence speed and overall model performance.

Sefik Emre Eskimez, You Zhang, Zhiyao Duan

Visual emotion expression plays an important role in audiovisual speech communication. In this work, we propose a novel approach to rendering visual emotion expression in speech-driven talking face generation. Specifically, we design an end-to-end talking face generation system that takes a speech utterance, a single face image, and a categorical emotion label as input to render a talking face video synchronized with the speech and expressing the conditioned emotion. Objective evaluation on image quality, audiovisual synchronization, and visual emotion expression shows that the proposed system outperforms a state-of-the-art baseline system. Subjective evaluation of visual emotion expression and video realness also demonstrates the superiority of the proposed system. Furthermore, we conduct a human emotion recognition pilot study using generated videos with mismatched emotions among the audio and visual modalities. Results show that humans respond to the visual modality more significantly than the audio modality on this task.

Dimitrios Dimitriadis, Kenichi Kumatani, Robert Gmyr et al.

In this paper, a Federated Learning (FL) simulation platform is introduced. The target scenario is Acoustic Model training based on this platform. To our knowledge, this is the first attempt to apply FL techniques to Speech Recognition tasks due to the inherent complexity. The proposed FL platform can support different tasks based on the adopted modular design. As part of the platform, a novel hierarchical optimization scheme and two gradient aggregation methods are proposed, leading to almost an order of magnitude improvement in training convergence speed compared to other distributed or FL training algorithms like BMUF and FedAvg. The hierarchical optimization offers additional flexibility in the training pipeline besides the enhanced convergence speed. On top of the hierarchical optimization, a dynamic gradient aggregation algorithm is proposed, based on a data-driven weight inference. This aggregation algorithm acts as a regularizer of the gradient quality. Finally, an unsupervised training pipeline tailored to FL is presented as a separate training scenario. The experimental validation of the proposed system is based on two tasks: first, the LibriSpeech task showing a speed-up of 7x and 6% Word Error Rate reduction (WERR) compared to the baseline results. The second task is based on session adaptation providing an improvement of 20% WERR over a competitive production-ready LAS model. The proposed Federated Learning system is shown to outperform the golden standard of distributed training in both convergence speed and overall model performance.

Sefik Emre Eskimez, Ross K Maddox, Chenliang Xu et al.

The presence of a corresponding talking face has been shown to significantly improve speech intelligibility in noisy conditions and for hearing impaired population. In this paper, we present a system that can generate landmark points of a talking face from an acoustic speech in real time. The system uses a long short-term memory (LSTM) network and is trained on frontal videos of 27 different speakers with automatically extracted face landmarks. After training, it can produce talking face landmarks from the acoustic speech of unseen speakers and utterances. The training phase contains three key steps. We first transform landmarks of the first video frame to pin the two eye points into two predefined locations and apply the same transformation on all of the following video frames. We then remove the identity information by transforming the landmarks into a mean face shape across the entire training dataset. Finally, we train an LSTM network that takes the first- and second-order temporal differences of the log-mel spectrogram as input to predict face landmarks in each frame. We evaluate our system using the mean-squared error (MSE) loss of landmarks of lips between predicted and ground-truth landmarks as well as their first- and second-order temporal differences. We further evaluate our system by conducting subjective tests, where the subjects try to distinguish the real and fake videos of talking face landmarks. Both tests show promising results.