Skip to main content
. 2024 Jul 4;67(11):4176–4185. doi: 10.1044/2024_JSLHR-24-00045

Figure 2.

Box plots depict the data for the word error rate in percentage by severity of speech impairment, mode of speech and ASR models. The severity level of speech impairment can be mild, moderate or severe. The mode of speech are Read and Conversational. The ASR models are personalized and unadapted. The description lists the minimum value, first quartile, median, third quartile, and the maximum value for each box. The data are as follows. 1. Personalized. Read mode. Mild: 2, 3, 3.5, 5, and 6. Moderate: 4, 5, 8, 10, and 12. Severe: 5, 9, 10, 13, and 16. Conversational mode. Mild: 1, 8, 10, 27, and 27. Moderate: 6, 13, 19, 26, and 36. Severe: 0, 18, 27, 35, and 40. 2. Unadapted. Read mode. Mild: 6, 15, 20, 25, and 26. Moderate: 27, 40, 44, 55, and 69. Severe: 34, 59, 74, 76, and 93. Conversational mode. Mild: 27, 34, 36, 40, and 40. Moderate: 36, 60, 68, 82, and 93. Severe: 36, 68, 91, 94, and 98.

Recognition accuracy of read and conversational speech as a function of speech impairment severity for the personalized and unadapted models. WER = word error rate.