No abstract

Continuous speech separation using a microphone array was shown to be promising in dealing with the speech overlap problem in natural conversation transcription. This paper proposes VarArray, an arraygeometry-agnostic speech separation neural network model. The proposed model is applicable to any number of microphones without retraining while leveraging the nonlinear correlation between the input channels. The proposed method adapts different elements that were proposed before separately, including transform-averageconcatenate, conformer speech separation, and inter-channel phase differences, and combines them in an efficient and cohesive way. Large-scale evaluation was performed with two real meeting transcription tasks by using a fully developed transcription system requiring no prior knowledge such as reference segmentations, which allowed us to measure the impact that the continuous speech separation system could have in realistic settings. The proposed model outperformed a previous approach to array-geometry-agnostic modeling for all of the geometry configurations considered, achieving asclite-based speaker-agnostic word error rates of 17.5% and 20.4% for the AMI development and evaluation sets, respectively, in the end-to-end setting using no ground-truth segmentations.

Language designers usually need to implement parsers and printers. Despite being two intimately related programs, in practice they are often designed separately, and then need to be revised and kept consistent as the language evolves. It will be more convenient if the parser and printer can be unified and developed in one single program, with their consistency guaranteed automatically. Furthermore, in certain scenarios (like showing compiler optimisation results to the programmer), it is desirable to have a more powerful reflective printer that, when an abstract syntax tree corresponding to a piece of program text is modified, can reflect the modification to the program text while preserving layouts, comments, and syntactic sugar. To address these needs, we propose a domain-specific language BIYACC, whose programs denote both a parser and a reflective printer for an unambiguous context-free grammar. BIYACC is based on the theory of bidirectional transformations, which helps to guarantee by construction that the pairs of parsers and reflective printers generated by BIYACC are consistent. We show that BIYACC is capable of facilitating many tasks such as Pombrio and Krishnamurthi's "resugaring", simple refactoring, and language evolution.

Language designers usually need to implement parsers and printers. Despite being two closely related programs, in practice they are often designed separately, and then need to be revised and kept consistent as the language evolves. It will be more convenient if the parser and printer can be unified and developed in a single program, with their consistency guaranteed automatically. Furthermore, in certain scenarios (like showing compiler optimisation results to the programmer), it is desirable to have a more powerful reflective printer that, when an abstract syntax tree corresponding to a piece of program text is modified, can propagate the modification to the program text while preserving layouts, comments, and syntactic sugar. To address these needs, we propose a domain-specific language BIYACC, whose programs denote both a parser and a reflective printer for a fully disambiguated context-free grammar. BIYACC is based on the theory of bidirectional transformations, which helps to guarantee by construction that the generated pairs of parsers and reflective printers are consistent. Handling grammatical ambiguity is particularly challenging: we propose an approach based on generalised parsing and disambiguation filters, which produce all the parse results and (try to) select the only correct one in the parsing direction; the filters are carefully bidirectionalised so that they also work in the printing direction and do not break the consistency between the parsers and reflective printers. We show that BIYACC is capable of facilitating many tasks such as Pombrio and Krishnamurthi's 'resugaring', simple refactoring, and language evolution.