word error rate
PulseAugur coverage of word error rate — every cluster mentioning word error rate across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New fine-tuning method enhances ASR post-editing with LLMs
Researchers have developed a novel method called oracle-guided fine-tuning to improve Automatic Speech Recognition (ASR) post-editing using large language models. This technique leverages N-best hypotheses to create cur…
-
New entity-aware metrics proposed for speech privacy evaluation
Researchers are proposing new metrics to evaluate privacy in speech processing, moving beyond traditional word error rates. These entity-aware metrics, adapted from Natural Language Processing, aim to better quantify in…
-
Itgan adapts Whisper for robust Arabic ASR at NADI 2026
Researchers from Itgan have detailed their systems for the NADI 2026 shared task, focusing on Arabic Automatic Speech Recognition (ASR) across robust, mixed-dialect, and code-switched scenarios. Their approach primarily…
-
AI voice assistants struggle with British accents, new benchmark reveals
A new benchmark, CavaBench, has been developed to evaluate the robustness of AI voice assistants to various British accents. Current systems, often trained on American English, exhibit significant word error rates (WER)…
-
New evaluation method reveals ASR/audio LM struggles with English-Yoruba code-switching
A new research paper introduces a switch-aware evaluation method for automatic speech recognition (ASR) and audio language models (audio LMs) when processing code-switched speech, specifically focusing on English and Yo…
-
New \"$\alpha$-split\" method enhances privacy for speech LLMs in federated learning
Researchers have developed a new method called \"$\alpha$-split\" to improve differential privacy in federated learning for multilingual speech large language models (speech-LLMs). Standard per-layer differential privac…
-
Whisper adaptation improves Greek song lyric transcription
Researchers have developed a new method for Automatic Lyric Transcription (ALT) specifically for Greek songs, addressing the challenges posed by melodic and rhythmic variations. By adapting OpenAI's Whisper model, they …
-
Whisper ASR tool aids Cantonese oral history transcription in New Zealand
A new paper explores the application of Automatic Speech Recognition (ASR) tools, specifically Whisper, for multilingual oral history research. The study focused on Cantonese revitalization efforts in New Zealand, findi…
-
New metric OVMI standardizes speech BCI evaluation
Researchers have introduced a new metric called open-vocabulary mutual information (OVMI) to standardize the evaluation of speech brain-computer interfaces (BCIs). This metric addresses the challenge of comparing BCIs t…
-
New batching method boosts speech transcription accuracy and speed
A new paper introduces Context-Aware Interleaved Batching, a method designed to improve the accuracy and efficiency of speech transcription. This technique addresses limitations in existing systems like WhisperX by main…
-
New VoiceCodeBench benchmark evaluates ASR accuracy for structured tokens
A new benchmark called VoiceCodeBench has been introduced to evaluate the accuracy of automatic speech recognition (ASR) systems in recovering specific structured tokens, such as identifiers and measured quantities. Tra…
-
Hugging Face adds Global South languages to Open ASR Leaderboard
Hugging Face has expanded its Open ASR Leaderboard by introducing two new evaluation sets, Monsoon en-IN and Monsoon hi-IN, to better represent languages from the Global South. These additions aim to address the known i…
-
New Mizo ASR system fine-tuned with Whisper and SraVaani models
Researchers have developed a new Automatic Speech Recognition (ASR) system for the Mizo language, a low-resource language, by collecting and curating over 17 hours of speech data. They fine-tuned three Whisper multiling…
-
AssemblyAI proposes Missed Entity Rate to improve speech-to-text accuracy
AssemblyAI argues that traditional word error rate (WER) metrics for speech-to-text systems are insufficient because they fail to account for the importance of specific entities like names, numbers, and medical terms. T…
-
Whisper ASR Models Adapted for Multilingual Medical Use
Researchers have analyzed how multilingual medical adaptation affects the internal representations of Whisper models. The study compared various fine-tuning strategies across different Whisper model sizes, finding that …
-
Custom speech recognition models rarely outperform universal AI, says AssemblyAI
AssemblyAI argues that custom speech recognition models are often unnecessary and can even be less accurate than modern universal models. While custom models were once the standard for improving accuracy in specialized …
-
AssemblyAI introduces cpWER to accurately measure speaker diarization accuracy
AssemblyAI has introduced a new metric called cpWER (concatenated minimum-permutation word error rate) to more accurately measure the performance of speaker diarization in speech-to-text systems. Unlike traditional word…
-
Vision-Language Models Show Subtle Hallucinations in Historical Document OCR
A new research paper analyzes the performance of vision-language models (VLMs) in transcribing historical documents, finding that while they outperform traditional optical character recognition (OCR) systems on standard…
-
Voice cloning enhances paralinguistic tasks and cross-lingual clinical speech analysis
A new research paper explores the use of voice cloning for data augmentation in paralinguistic tasks, particularly for clinical applications where labeled data is scarce. The study benchmarks eight voice cloning models,…
-
Voice cloning models preserve paralinguistic signals for clinical speech tasks
Researchers have evaluated eight voice cloning models to determine their effectiveness in preserving paralinguistic signals for speech tasks, particularly in clinical settings where labeled data is scarce. The study fou…