Interspeech 2022

1,122 papers
Interspeech 2022 featured research focused on end-to-end learning from raw audio waveforms and improving robust speech processing in out-of-distribution environments. Technical developments emphasized architectural innovations, such as asymmetric convolutions for time-frequency processing and the integration of local and global contexts. Multi-resolution architectures and precise temporal grounding for sound event detection were also prominent themes during this edition. The year's contributions addressed diverse applications, including the automatic detection of speech sound disorders in children and privacy-preserving techniques for automatic speech recognition using self-supervised learning. Research also explored speaker identification challenges, such as mitigating within-household confusability through graph-based multi-view fusion. Other work included advancements in speech enhancement via cross-layer similarity knowledge distillation and any-to-any voice conversion using contrastive predictive coding frameworks. Prolific contributors to this edition included researchers such as Chi-Chun Lee, Yu Tsao, and Hsin-Min Wang.

Top Research Directions

Research directions most represented by Interspeech 2022 papers.

Learning High-Level Acoustic Structures from Unlabeled Audio 41
Overcoming Data Scarcity in Low-Resource Speech Recognition 28
Integrating Local and Global Context in Speech Architectures 27
Balancing Latency and Accuracy in Streaming Speech Recognition 25
Partitioning Variance with Linear Mixed Effects Models 21
Generalization to Unseen Acoustic Environments 20
Zero Shot Speaker Generalization for Speech Synthesis 20
The Perceptual Evaluation Gap in Generative Audio 20
Joint Optimization of Speech Enhancement and Recognition 18
Fine-Grained Prosodic Control for Expressive Speech Synthesis 18
The Single Speaker Generalization Bottleneck in Speech Synthesis 18
Complex Spectral Mapping for Phase Aware Signal Reconstruction 17
Robust Natural Language Understanding for Noisy Speech Transcripts 17
Resolving Overlapping Speech in Multi-Speaker Audio 17
Parameter Efficient Speech and Audio Processing 16

Top Authors

AuthorPapers
Shinji Watanabe23
Helen M. Meng14
Hung-yi Lee10
Zhiyong Wu10
Joon-Hyuk Chang9
Yanmin Qian9
Yonghong Yan9
Haizhou Li8
Jinyu Li8
Longbiao Wang8
Jing Xiao7
Lei Xie7
Yu Tsao7
Alan Black6
Jesus Villalba6
Joseph Keshet6
Jun Du6
Michael Auli6
Najim Dehak6
Yu Wu6

Impactful Papers

Papers from Interspeech 2022 that shaped research clusters.

Foundational

Influential

Papers

Title
Streaming parallel transducer beam search with fast slow cascaded encoders
Streaming Speaker-Attributed ASR with Token-Level Speaker Embeddings
Streaming Target-Speaker ASR with Neural Transducer
STUDIES: Corpus of Japanese Empathetic Dialogue Speech Towards Friendly Voice Agent
Sub-8-Bit Quantization Aware Training for 8-Bit Neural Network Accelerator with On-Device Speech Recognition
Supervision-Guided Codebooks for Masked Prediction in Speech Pre-training
SVTS: Scalable Video-to-Speech Synthesis
Syllable sequence of /a/+/ta/ can be heard as /atta/ in Japanese with visual or tactile cues
Synthesizing Near Native-accented Speech for a Non-native Speaker by Imitating the Pronunciation and Prosody of a Native Speaker
TALCS: An open-source Mandarin-English code-switching corpus and a speech recognition baseline
Tandem Multitask Training of Speaker Diarisation and Speech Recognition for Meeting Transcription
Target Confusion in End-to-end Speaker Extraction: Analysis and Approaches
TaylorBeamformer: Learning All-Neural Beamformer for Multi-Channel Speech Enhancement from Taylor's Approximation Theory
TB or not TB? Acoustic cough analysis for tuberculosis classification
Telling self-defining memories: An acoustic study of natural emotional speech productions
Temporal coding with magnitude-phase regularization for sound event detection
Temporal Self Attention-Based Residual Network for Environmental Sound Classification
Text aware Emotional Text-to-speech with BERT
Text-driven Emotional Style Control and Cross-speaker Style Transfer in Neural TTS
Text-Driven Separation of Arbitrary Sounds
Text-Only Domain Adaptation Based on Intermediate CTC
Text-to-speech synthesis using spectral modeling based on non-negative autoencoder
The 1st Clarity Prediction Challenge: A machine learning challenge for hearing aid intelligibility prediction
The CLIPS System for 2022 Spoofing-Aware Speaker Verification Challenge
The discrimination of [zi]-[dʑi] by Japanese listeners and the prospective phonologization of /zi/
The DKU-OPPO System for the 2022 Spoofing-Aware Speaker Verification Challenge
The Effectiveness of Time Stretching for Enhancing Dysarthric Speech for Improved Dysarthric Speech Recognition
The effect of backward noise on lexical tone discrimination in Mandarin-speaking amusics
The effect of increasing acoustic and linguistic complexity on auditory processing: an EEG study
The Effects of Implicit and Explicit Feedback in an ASR-based Reading Tutor for Dutch First-graders
The Emotion is Not One-hot Encoding: Learning with Grayscale Label for Emotion Recognition in Conversation
The HCCL System for the NIST SRE21
The Magnitude and Phase based Speech Representation Learning using Autoencoder for Classifying Speech Emotions using Deep Canonical Correlation Analysis
The mapping between syntactic and prosodic phrasing in English and Mandarin
The Prosody of Cheering in Sport Events
The THUEE System Description for the IARPA OpenASR21 Challenge
The VoiceMOS Challenge 2022
The ZevoMOS entry to VoiceMOS Challenge 2022
Three-dimensional finite-difference time-domain acoustic analysis of simplified vocal tract shapes
Thutmose Tagger: Single-pass neural model for Inverse Text Normalization
Time-domain Ad-hoc Array Speech Enhancement Using a Triple-path Network
Tiny-Sepformer: A Tiny Time-Domain Transformer Network For Speech Separation
TMGAN-PLC: Audio Packet Loss Concealment using Temporal Memory Generative Adversarial Network
Token-level Speaker Change Detection Using Speaker Difference and Speech Content via Continuous Integrate-and-fire
Tokenwise Contrastive Pretraining for Finer Speech-to-BERT Alignment in End-to-End Speech-to-Intent Systems
TopicKS: Topic-driven Knowledge Selection for Knowledge-grounded Dialogue Generation
Toward Corpus Size Requirements for Training and Evaluating Depression Risk Models Using Spoken Language
Toward Fairness in Speech Recognition: Discovery and mitigation of performance disparities
Toward Low-Cost End-to-End Spoken Language Understanding
Towards Automated Counselling Decision-Making: Remarks on Therapist Action Forecasting on the AnnoMI Dataset

« prev Page 20/23 next »