Recurrent Neural Aligner: An Encoder-Decoder Neural Network Model for Sequence to Sequence Mapping

By Hasim Sak, Matt Shannon, Kanishka Rao, Franccoise Beaufays. 2017

136 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Natural Language Processing, Model Architectures

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Model Architectures

Citations in the corpus

State-of-the-Art Speech Recognition with Sequence-to-Sequence Models (2017, methodology): “There have been a variety of methods explored in the literature to address this issue in the context of sequence-to-sequence models [22, 23, 5, 12].”
Exploring architectures, data and units for streaming end-to-end speech recognition with RNN-transducer (2017, background): “Several streaming encoder-decoder architectures have been proposed previously, including the neural transducer [11], the recurrent neural aligner (RNA) [12], and the”
Knowledge Distillation for Sequence Model (2018, unknown): “Future works include extending the sequence-level knowledge distillation to several newlyproposed sequence criteria, e.g. RNN transducer [27], recurrent neural aligner (RNA) [28], neural segmental model [29] and attention based encoder-decoder [30].”
Sequence-to-sequence Modelling of F0 for Speech Emotion Conversion (2019, background): “Attention mechanisms augment the performance of the basic seq2seq model by allowing the implicit alignment of input and output sequences, and are the standard today in seq2seq architectures [25, 26], with some recent applications to speech processing [27].”
Improving Attention-Based End-to-End ASR Systems with Sequence-Based Loss Functions (2018, background): “Some followup work have integrated character dependencies directly into the model, such as recurrent neural network transducer (RNN-transducer) [22, 20] and recurrent neural aligner [23, 24].”
Recent Advances in End-to-End Automatic Speech Recognition (2021, methodology): “In addition to these three popular E2E models, there are also other E2E models such as neural segmental model [80] and recurrent neural aligner [81], to name a few.”
Incremental Text to Speech for Neural Sequence-to-Sequence Models using Reinforcement Learning (2020, methodology): “The development of these streaming, end-to-end architectures has seen considerable attention for the tasks of automatic speech recognition [7, 8, 9, 10] and machine translation (MT) [11, 12, 13, 14].”
Recent progresses in deep learning based acoustic models (2017, methodology): “On the other hand, RNN transducer [76] and RNN aligner [77] extend CTC modeling by changing the objective function and the training process to remove the frame-independence assumption of CTC. Specif-6 ically, RNN transducer was shown very effective [78], [79] as it incorporates acoustic model with…”
Improving Unsupervised Style Transfer in end-to-end Speech Synthesis with end-to-end Speech Recognition (2018, methodology): “The TTS model we used here is GST-tacotron [8], which is already introduced in Section 3, while ASR model can be any end-toend seq2seq model[28, 29, 30, 31, 32, 33, 34].”
End-to-End Speech Recognition: A Survey (2023, background): “…of non-blank symbols in the partial alignment sequence ( a 1 , . . . , a t − 1 ) , and q t − 1 = NN ( ·| a t − 1 , . . . , a 1 ) represents the output of a neural network which summarizes the entire partial alignment sequence, where NN ( · ) represents a suitable neural network (an LSTM in [46]).”
Extending Recurrent Neural Aligner for Streaming End-to-End Speech Recognition in Mandarin (2018, methodology): “Recently, a considerable amount of works have demonstrated the simpliﬁcation and effectiveness of end-to-end models in the ASR ﬁeld [1, 2, 3, 4, 5].”
Streaming small-footprint keyword spotting using sequence-to-sequence models (2017, background): “Examples of such models include the recurrent neural network transducer (RNNT) [16, 17], the recurrent neural aligner [18], connectionist temporal classification (CTC) [19] with grapheme [20, 21], syllable [22] or word targets [23], and attention-based models [24, 25, 26].”
Minimum Bayes Risk Training of RNN-Transducer for End-to-End Speech Recognition (2019, methodology): “Sampling-based discriminative training technique was also proposed [16, 21] to fine-tune a RNA model to minimize expected word error rate (WER).”
Self-attention Aligner: A Latency-control End-to-end Model for ASR Using Self-attention Network and Chunk-hopping (2019, methodology): “Self-attention aligner (SAA) is a RNN-free end-to-end model that contains three parts: an encoder, a decoder and a CTC-like alignment mechanism.”