Advances in Robust Speech Recognition

The field of speech recognition is moving towards developing more robust and trustworthy models, with a focus on improving performance in noisy environments and handling diverse user groups. Recent research has highlighted the importance of internal consistency in models, with techniques such as multi-granularity consistency frameworks showing promising results. Additionally, there is a growing interest in developing models that can handle out-of-vocabulary words, named entity correction, and cross-lingual phoneme recognition. Noteworthy papers include: MGSC, which introduces a model-agnostic framework for enforcing internal self-consistency, and Whisper based Cross-Lingual Phoneme Recognition, which proposes a novel bilingual speech recognition approach for Vietnamese and English. Other notable papers include Attention2Probability, which proposes attention-driven terminology probability estimation, and ReSURE, which introduces an adaptive learning method for regularizing supervision unreliability in multi-turn dialogue fine-tuning.

Sources

MGSC: A Multi-granularity Consistency Framework for Robust End-to-end Asr

Benchmarking Training Paradigms, Dataset Composition, and Model Scaling for Child ASR in ESPnet

Talking to Robots: A Practical Examination of Speech Foundation Models for HRI Applications

Zero-shot Context Biasing with Trie-based Decoding using Synthetic Multi-Pronunciation

H-PRM: A Pluggable Hotword Pre-Retrieval Module for Various Speech Recognition Systems

Attention2Probability: Attention-Driven Terminology Probability Estimation for Robust Speech-to-Text System

Whisper based Cross-Lingual Phoneme Recognition between Vietnamese and English

CAM\~OES: A Comprehensive Automatic Speech Recognition Benchmark for European Portuguese

ReSURE: Regularizing Supervision Unreliability for Multi-turn Dialogue Fine-tuning

Generative Annotation for ASR Named Entity Correction

OLMoASR: Open Models and Data for Training Robust Speech Recognition Models

Built with on top of