Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

By Sharon Goldwater, Dan Jurafsky, Christopher D. Manning. 2010

204 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Experimental and Cognitive Psychology, Speech & Audio

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

1. Hung-yi Lee
2. James R. Glass
3. Yonatan Belinkov
4. Lin-Shan Lee
5. Dan Jurafsky
6. Percy Liang
7. Mohit Bansal
8. Noah A. Smith

In Experimental and Cognitive Psychology

1. Rachael Tatman
2. James R. Glass
3. Hung-yi Lee
4. Dan Jurafsky
5. J. Rickford
6. Andrew Y. Ng

In Speech & Audio

1. Hung-yi Lee
2. Lin-Shan Lee
3. James R. Glass
4. Chun-an Chan
5. Rachael Tatman
6. Andrew Y. Ng
7. Andrew L. Maas
8. Ziang Xie

Citations in the corpus

• Racial disparities in automated speech recognition (2020, background): “Past work has also found that ASRs perform somewhat worse on conversational speech from male speakers than female speakers, likely due to male speakers using more informal style with shorter, more reduced pronunciations and more disfluencies (28, 29).”
• Gender and Dialect Bias in YouTube’s Automatic Captions (2017, background): “Pitch differences are one of the most reliable and well-studied perceptual markers of gender in speech (Wu and Childers, 1991; Gelfer and Mikos, 2005) and speech with a high fundamental frequency (typical of women’s speech) has also been found to be more difficult for automatic speech recognizers (H”
• Lexicon-Free Conversational Speech Recognition with Neural Networks (2015, methodology): “We find that OOVs tend to trigger clusters of errors in the HMM-GMM system, an observation that has been systematically explored in previous work (Goldwater et al., 2010).”
• Spoken Content Retrieval—Beyond Cascading Speech Recognition with Text Retrieval (2015, background): “usually cause more ASR errors [163], thus helpful to identify the false alarms.”
• Effects of Talker Dialect, Gender & Race on Accuracy of Bing Speech and YouTube Automatic Captions (2017, result): “Difference in Accuracy by Talker Gender Previous studies have found differences in ASR accuracy among genders [16, 17, 18], and earlier work found that men’s voices were recognized with more accuracy by YouTube’s automatic captions [1].”