Personalized speech enhancement (PSE) models can improve the audio quality of teleconferencing systems by adapting to the characteristics of a speaker's voice.However, most existing methods require a separate speaker embedding model to extract a vector representation of the speaker from enrollment audio, which adds complexity to the training and deployment process.We propose to use the internal representation of the PSE model itself as the speaker embedding, thereby avoiding the need for a separate model.We show that our approach performs equally well or better than the standard method of using a pre-trained speaker embedding model on noise suppression and echo cancellation tasks.Moreover, our approach surpasses the ICASSP 2023 Deep Noise Suppression Challenge winner by 0.15 in Mean Opinion Score.

The ICASSP 2022 Acoustic Echo Cancellation Challenge is intended to stimulate research in acoustic echo cancellation (AEC), which is an important area of speech enhancement and still a top issue in audio communication. This is the third AEC challenge and it is enhanced by including mobile scenarios, adding speech recognition word accuracy rate as a metric, and making the audio 48 kHz. We open source two large datasets to train AEC models under both single talk and double talk scenarios. These datasets consist of recordings from more than 10,000 real audio devices and human speakers in real environments, as well as a synthetic dataset. We open source an online subjective test framework and provide an online objective metric service for researchers to quickly test their results. The winners of this challenge were selected based on the average Mean Opinion Score (MOS) achieved across all scenarios and the word accuracy rate.

Traditionally, the quality of acoustic echo cancellers is evaluated using intrusive speech quality assessment measures such as ERLE [1] and PESQ [2], or by carrying out subjective laboratory tests [3], [4]. Unfortunately, the former are not well correlated with human subjective measures, while the latter are time and resource consuming to carry out [5]. We provide a new tool for speech quality assessment for echo impairment which can be used to evaluate the performance of acoustic echo cancellers. More precisely, we develop a neural network model to evaluate call quality degradations in two separate categories: echo and degradations from other sources. We show that our model is accurate as measured by correlation with human subjective quality ratings. Our tool can be used effectively to stack rank echo cancellation models. AECMOS is being made publicly available as an Azure service.

Audio Packet Loss Concealment (PLC) is the hiding of gaps in audio streams caused by data transmission failures in packet switched networks.This is a common problem, and of increasing importance as end-to-end VoIP telephony and teleconference systems become the default and ever more widely used form of communication in business as well as in personal usage.This paper presents the INTERSPEECH 2022 Audio Deep Packet Loss Concealment challenge.We first give an overview of the PLC problem, and introduce some classical approaches to PLC as well as recent work.We then present the open source dataset released as part of this challenge as well as the evaluation methods and metrics used to determine the winner.We also briefly introduce PLCMOS, a novel data-driven metric that can be used to quickly evaluate the performance PLC systems.Finally, we present the results of the INTERSPEECH 2022 Audio Deep PLC Challenge, and provide a summary of important takeaways.

The quality of acoustic echo cancellers (AECs) in real-time communication systems is typically evaluated using objective metrics like ERLE [1] and PESQ [2], and less commonly with lab-based subjective tests like ITU-T Rec. P.831 [3]. We will show that these objective measures are not well correlated to subjective measures. We then introduce an open-source crowdsourcing approach for subjective evaluation of echo impairment which can be used to evaluate the performance of AECs. We provide a study that shows this tool is highly reproducible. This new tool has been recently used in the ICASSP 2021 AEC Challenge [4] which made the challenge possible to do quickly and cost effectively.

The ICASSP 2021 Acoustic Echo Cancellation Challenge is intended to stimulate research in the area of acoustic echo cancellation (AEC), which is an important part of speech enhancement and still a top issue in audio communication and conferencing systems. Many recent AEC studies report good performance on synthetic datasets where the train and test samples come from the same underlying distribution. However, the AEC performance often degrades significantly on real recordings. Also, most of the conventional objective metrics such as echo return loss enhancement (ERLE) and perceptual evaluation of speech quality (PESQ) do not correlate well with subjective speech quality tests in the presence of background noise and reverberation found in realistic environments. In this challenge, we open source two large datasets to train AEC models under both single talk and double talk scenarios. These datasets consist of recordings from more than 2,500 real audio devices and human speakers in real environments, as well as a synthetic dataset. We open source two large test sets, and we open source an online subjective test framework for researchers to quickly test their results. The winners of this challenge will be selected based on the average Mean Opinion Score (MOS) achieved across all different single talk and double talk scenarios.

We show that a wide class of bidirectional data-flow analyses and program optimizations based on them admit declarative descriptions in the form of type systems. The salient feature is a clear separation between what constitutes a valid analysis and how the strongest one can be computed (via the type checking versus principal type inference distinction). The approach also facilitates elegant relational semantic soundness definitions and proofs for analyses and optimizations, with an application to mechanical transformation of program proofs, useful in proof-carrying code. Unidirectional forward and backward analyses are covered as special cases; the technicalities in the general bidirectional case arise from more subtle notions of valid and principal types. To demonstrate the viability of the approach we consider two examples that are inherently bidirectional: type inference (seen as a data-flow problem) for a structured language where the type of a variable may change over a program's run and the analysis underlying a stack usage optimization for a stack-based low-level language.

We present a new, semantics-based approach to mechanical program repair where the intended meaning of broken programs (i.e., programs that may abort under a given, error-admitting language semantics) can be defined by a special, error-compensating semantics. Program repair can then become a compile-time, mechanical program transformation based on a program analysis. It turns a given program into one whose evaluations under the error-admitting semantics agree with those of the given program under the error-compensating semantics. We present the analysis and transformation as a type system with a transformation component, following the type-systematic approach to program optimization from our earlier work. The type-systematic method allows for simple soundness proofs of the repairs, based on a relational interpretation of the type system, as well as mechanical transformability of program correctness proofs between the Hoare logics for the error-compensating and error-admitting semantics. We first demonstrate our approach on the repair of file-handling programs with missing or superfluous open and close statements. Our framework shows that this repair is strikingly similar to partial redundancy elimination optimization commonly used by compilers. In a second example, we demonstrate the repair of programs operating a queue that can over- and underflow, including mechanical transformation of program correctness proofs.

Partial redundancy elimination is a subtle optimization which performs common subexpression elimination and expression motion at the same time. In this paper, we use it as an example to promote and demonstrate the scalability of the technology of proof optimization. By this we mean automatic transformation of a given program's Hoare logic proof of functional correctness or resource usage into one of the optimized program, guided by a type-derivation representation of the result of the underlying dataflow analyses. A proof optimizer is a useful tool for the producer's side in a natural proof-carrying code scenario where programs are proved correct prior to optimizing compilation before transmission to the consumer.

Data-flow analyses, such as live variables analysis, available expressions analysis etc., are usefully specifiable as type systems. These are sound and, in the case of distributive analysis frameworks, complete wrt. appropriate natural semantics on abstract properties. Applications include certification of analyses and "optimization" of functional correctness proofs alongside programs. On the example of live variables analysis, we show that analysis type systems are applied versions of more foundational Hoare logics describing either the same abstract property semantics as the type system (liveness states) or a more concrete natural semantics on transition traces of a suitable kind (future defs and uses). The rules of the type system are derivable in the Hoare logic for the abstract property semantics and those in turn in the Hoare logic for the transition trace semantics. This reduction of the burden of trusting the certification vehicle can be compared to foundational proof-carrying code, where general-purpose program logics are preferred to special-purpose type systems and universal logic to program logics. We also look at conditional liveness analysis to see that the same foundational development is also possible for conditional data-flow analyses proceeding from type systems for combined "standard state and abstract property" semantics.

A way of combining object-oriented and structural paradigms of software composition is demonstrated in a tool for generative programming. Metaclasses are introduced that are components with specifications called metainterfaces. Automatic code generation is used that is based on structural synthesis of programs. This guarantees that problems of handling data dependencies, order of application of components, usage of higher-order control structures etc are handled automatically. Specifications can be written either in a specification language or given visually on an architectural level. The tool is Java-based and portable.