Beamforming is a powerful tool designed to enhance speech signals from the direction of a target source. Computing the beamforming filter requires estimating spatial covariance matrices (SCMs) of the source and noise signals. Time-frequency masks are often used to compute these SCMs. Most studies of mask-based beamforming have assumed that the sources do not move. However, sources often move in practice, which causes performance degradation. In this paper, we address the problem of mask-based beamforming for moving sources. We first review classical approaches to tracking a moving source, which perform online or blockwise computation of the SCMs. We show that these approaches can be interpreted as computing a sum of instantaneous SCMs weighted by attention weights. These weights indicate which time frames of the signal to consider in the SCM computation. Online or blockwise computation assumes a heuristic and deterministic way of computing these attention weights that, although simple, may not result in optimal performance. We thus introduce a learning-based framework that computes optimal attention weights for beamforming. We achieve this using a neural network implemented with self-attention layers. We show experimentally that our proposed framework can greatly improve beamforming performance in moving source situations while maintaining high performance in non-moving situations, thus enabling the development of mask-based beamformers robust to source movements.
Target speech extraction (TSE) extracts the speech of a target speaker in a mixture given auxiliary clues characterizing the speaker, such as an enrollment utterance. TSE addresses thus the challenging problem of simultaneously performing separation and speaker identification. There has been much progress in extraction performance following the recent development of neural networks for speech enhancement and separation. Most studies have focused on processing mixtures where the target speaker is actively speaking. However, the target speaker is sometimes silent in practice, i.e., inactive speaker (IS). A typical TSE system will tend to output a signal in IS cases, causing false alarms. This is a severe problem for the practical deployment of TSE systems. This paper aims at understanding better how well TSE systems can handle IS cases. We consider two approaches to deal with IS, (1) training a system to directly output zero signals or (2) detecting IS with an extra speaker verification module. We perform an extensive experimental comparison of these schemes in terms of extraction performance and IS detection using the LibriMix dataset and reveal their pros and cons.
In many situations, we would like to hear desired sound events (SEs) while being able to ignore interference. Target sound extraction (TSE) aims at tackling this problem by estimating the sound of target SE classes in a mixture while suppressing all other sounds. We can achieve this with a neural network that extracts the target SEs by conditioning it on clues representing the target SE classes. Two types of clues have been proposed, i.e., target SE class labels and enrollment sound samples similar to the target sound. Systems based on SE class labels can directly optimize embedding vectors representing the SE classes, resulting in high extraction performance. However, extending these systems to the extraction of new SE classes not encountered during training is not easy. Enrollment-based approaches extract SEs by finding sounds in the mixtures that share similar characteristics to the enrollment. These approaches do not explicitly rely on SE class definitions and can thus handle new SE classes. In this paper, we introduce a TSE framework, SoundBeam, that combines the advantages of both approaches. We also perform an extensive evaluation of the different TSE schemes using synthesized and real mixtures, which shows the potential of SoundBeam.
Speaker diarization has been investigated extensively as an important central task for meeting analysis. Recent trend shows that integration of end-to-end neural (EEND)-and clustering-based diarization is a promising approach to handle realistic conversational data containing overlapped speech with an arbitrarily large number of speakers, and achieved state-of-the-art results on various tasks. However, the approaches proposed so far have not realized {\it tight} integration yet, because the clustering employed therein was not optimal in any sense for clustering the speaker embeddings estimated by the EEND module. To address this problem, this paper introduces a {\it trainable} clustering algorithm into the integration framework, by deep-unfolding a non-parametric Bayesian model called the infinite Gaussian mixture model (iGMM). Specifically, the speaker embeddings are optimized during training such that it better fits iGMM clustering, based on a novel clustering loss based on Adjusted Rand Index (ARI). Experimental results based on CALLHOME data show that the proposed approach outperforms the conventional approach in terms of diarization error rate (DER), especially by substantially reducing speaker confusion errors, that indeed reflects the effectiveness of the proposed iGMM integration.
It is challenging to improve automatic speech recognition (ASR) performance in noisy conditions with single-channel speech enhancement (SE). In this paper, we investigate the causes of ASR performance degradation by decomposing the SE errors using orthogonal projection-based decomposition (OPD). OPD decomposes the SE errors into noise and artifact components. The artifact component is defined as the SE error signal that cannot be represented as a linear combination of speech and noise sources. We propose manually scaling the error components to analyze their impact on ASR. We experimentally identify the artifact component as the main cause of performance degradation, and we find that mitigating the artifact can greatly improve ASR performance. Furthermore, we demonstrate that the simple observation adding (OA) technique (i.e., adding a scaled version of the observed signal to the enhanced speech) can monotonically increase the signal-to-artifact ratio under a mild condition. Accordingly, we experimentally confirm that OA improves ASR performance for both simulated and real recordings. The findings of this paper provide a better understanding of the influence of SE errors on ASR and open the door to future research on novel approaches for designing effective single-channel SE front-ends for ASR.
The combination of a deep neural network (DNN) -based speech enhancement (SE) front-end and an automatic speech recognition (ASR) back-end is a widely used approach to implement overlapping speech recognition. However, the SE front-end generates processing artifacts that can degrade the ASR performance. We previously found that such performance degradation can occur even under fully overlapping conditions, depending on the signal-to-interference ratio (SIR) and signal-to-noise ratio (SNR). To mitigate the degradation, we introduced a rule-based method to switch the ASR input between the enhanced and observed signals, which showed promising results. However, the rule's optimality was unclear because it was heuristically designed and based only on SIR and SNR values. In this work, we propose a DNN-based switching method that directly estimates whether ASR will perform better on the enhanced or observed signals. We also introduce soft-switching that computes a weighted sum of the enhanced and observed signals for ASR input, with weights given by the switching model's output posteriors. The proposed learning-based switching showed performance comparable to that of rule-based oracle switching. The soft-switching further improved the ASR performance and achieved a relative character error rate reduction of up to 23 % as compared with the conventional method.
Speech summarization, which generates a text summary from speech, can be achieved by combining automatic speech recognition (ASR) and text summarization (TS). With this cascade approach, we can exploit state-of-the-art models and large training datasets for both subtasks, i.e., Transformer for ASR and Bidirectional Encoder Representations from Transformers (BERT) for TS. However, ASR errors directly affect the quality of the output summary in the cascade approach. We propose a cascade speech summarization model that is robust to ASR errors and that exploits multiple hypotheses generated by ASR to attenuate the effect of ASR errors on the summary. We investigate several schemes to combine ASR hypotheses. First, we propose using the sum of sub-word embedding vectors weighted by their posterior values provided by an ASR system as an input to a BERT-based TS system. Then, we introduce a more general scheme that uses an attention-based fusion module added to a pre-trained BERT module to align and combine several ASR hypotheses. Finally, we perform speech summarization experiments on the How2 dataset and a newly assembled TED-based dataset that we will release with this paper. These experiments show that retraining the BERT-based TS system with these schemes can improve summarization performance and that the attention-based fusion module is particularly effective.
In typical multi-talker speech recognition systems, a neural network-based acoustic model predicts senone state posteriors for each speaker. These are later used by a single-talker decoder which is applied on each speaker-specific output stream separately. In this work, we argue that such a scheme is sub-optimal and propose a principled solution that decodes all speakers jointly. We modify the acoustic model to predict joint state posteriors for all speakers, enabling the network to express uncertainty about the attribution of parts of the speech signal to the speakers. We employ a joint decoder that can make use of this uncertainty together with higher-level language information. For this, we revisit decoding algorithms used in factorial generative models in early multi-talker speech recognition systems. In contrast with these early works, we replace the GMM acoustic model with DNN, which provides greater modeling power and simplifies part of the inference. We demonstrate the advantage of joint decoding in proof of concept experiments on a mixed-TIDIGITS dataset.
Many state-of-the-art neural network-based source separation systems use the averaged Signal-to-Distortion Ratio (SDR) as a training objective function. The basic SDR is, however, undefined if the network reconstructs the reference signal perfectly or if the reference signal contains silence, e.g., when a two-output separator processes a single-speaker recording. Many modifications to the plain SDR have been proposed that trade-off between making the loss more robust and distorting its value. We propose to switch from a mean over the SDRs of each individual output channel to a global SDR over all output channels at the same time, which we call source-aggregated SDR (SA-SDR). This makes the loss robust against silence and perfect reconstruction as long as at least one reference signal is not silent. We experimentally show that our proposed SA-SDR is more stable and preferable over other well-known modifications when processing meeting-style data that typically contains many silent or single-speaker regions.
Automatic transcription of meetings requires handling of overlapped speech, which calls for continuous speech separation (CSS) systems. The uPIT criterion was proposed for utterance-level separation with neural networks and introduces the constraint that the total number of speakers must not exceed the number of output channels. When processing meeting-like data in a segment-wise manner, i.e., by separating overlapping segments independently and stitching adjacent segments to continuous output streams, this constraint has to be fulfilled for any segment. In this contribution, we show that this constraint can be significantly relaxed. We propose a novel graph-based PIT criterion, which casts the assignment of utterances to output channels in a graph coloring problem. It only requires that the number of concurrently active speakers must not exceed the number of output channels. As a consequence, the system can process an arbitrary number of speakers and arbitrarily long segments and thus can handle more diverse scenarios. Further, the stitching algorithm for obtaining a consistent output order in neighboring segments is of less importance and can even be eliminated completely, not the least reducing the computational effort. Experiments on meeting-style WSJ data show improvements in recognition performance over using the uPIT criterion.