Data collection of Japanese dialects and its influence into speech recognition

This paper outlines the background development of “intelligent” technologies such as speech recognition. Despite significant progress in the development of these technologies, they still fall short in many areas, and rapid advances in areas such as dictation are actually stalled. In this paper we have proposed semi-automatic solutions — smart integration of human and intelligent efforts. One such technique involves improvement to the speech recognition editing interface, thereby reducing the perception of errors to the viewer. Other techniques that are described in the paper are batch enrollment, which allows the user to reduce the amount of time required for enrollment, and content spotting, which can be used for applications that have repeated content flow, such as movies or museum tours. The paper also suggests a general concept of distributive training of speech recognition systems that is based on data collection across a network.

Download Full-text

Automatic Speech Recognition Applications: A Study of Methods for Defining Command Vocabularies

Proceedings of the Human Factors and Ergonomics Society Annual Meeting ◽

10.1177/154193129503900307 ◽

1995 ◽

Vol 39 (3) ◽

pp. 203-207

Author(s):

Roger B. Garberg

Keyword(s):

Speech Recognition ◽

Data Collection ◽

Automatic Speech Recognition ◽

Service Operations ◽

Critical Dimension ◽

Data Collection And Analysis ◽

Cognitive Activities ◽

Time Required ◽

Request Service ◽

Divide Attention

Phoneme-based automatic speech recognition (ASR) technology enables designers to easily create custom command words or phrases that users can employ to request service operations. In this paper, I report results from two experiments concerning important dimensions of these ASR command vocabularies, including command naturalness/appropriateness and command recallability. Ease of recall is a critical dimension for assessing ASR commands used in multi-step applications since service subscribers may be engaged in several different cognitive activities that divide attention. Yet techniques for measuring command recallability can be difficult to implement owing to the time required for data collection and analysis. Results of these studies indicate the the dimension of command “naturalness” and memorability are closely related: under appropriate conditions, the simple procedures associated with measuring command naturalness or appropriateness can predict retrievability of command expressions.

Download Full-text

On EMG Silent-speech Recognition Data Collection

2019 International Multi-Conference on Engineering, Computer and Information Sciences (SIBIRCON) ◽

10.1109/sibircon48586.2019.8958407 ◽

2019 ◽

Author(s):

Andrey Zubkov ◽

Andrey Marinov

Keyword(s):

Speech Recognition ◽

Data Collection ◽

Silent Speech

Download Full-text

Pengembangan Perangkat Speech Recognition Bagi Penyandang Bisu Tuli Sebagai Sarana Komunikasi

Jurnal MediaTIK ◽

10.26858/jmtik.v3i1.14388 ◽

2020 ◽

Vol 3 (1) ◽

pp. 49

Author(s):

Umrah - Sahni ◽

Udin Sidik Sidin ◽

Muhammad - Riska

Keyword(s):

Speech Recognition ◽

Data Collection ◽

Response Rate ◽

Descriptive Analysis ◽

Response Category ◽

Average Response ◽

Average Response Rate ◽

The Public ◽

South Sulawesi

This study aims to produce a design for the development of speech recognition devices for deaf-mute people to make it easier for the public to communicate with deaf mute visitors during emergencies by using a sound sensor that is speech recognition. Data collection techniques used were observation, questionnaires and documentation. Then the data collected will be analyzed using descriptive analysis. The results of the study were obtained from 25 respondents in the SLB of South Sulawesi Province Development, resulting in an overall average response rate of 80% from the highest value of 100% and the lowest value of 0%. This tool works with voice commands that have been done by the word train first first Then if the word is detected, the word will automatically appear on the LCD screen and simultaneously with the vibration produced by the vibrator. So it can be concluded that the speech recognition device for deaf mutes is good based on the respondent response category table.

Download Full-text

A Comparison of Two Word-Recognition Tasks in Multitalker Babble: Speech Recognition in Noise Test (SPRINT) and Words-in-Noise Test (WIN)

Journal of the American Academy of Audiology ◽

10.3766/jaaa.19.7.4 ◽

2008 ◽

Vol 19 (07) ◽

pp. 548-556 ◽

Cited By ~ 13

Author(s):

Richard H. Wilson ◽

Wendy B. Cates

Keyword(s):

Hearing Loss ◽

Speech Recognition ◽

Data Collection ◽

Word Recognition ◽

Concurrent Validity ◽

Pure Tone ◽

Normal Hearing ◽

Pure Tone Average ◽

Noise Test ◽

Speech Recognition In Noise

Background: The Speech Recognition in Noise Test (SPRINT) is a word-recognition instrument that presents the 200 Northwestern University Auditory Test No. 6 (NU-6) words binaurally at 50 dB HL in a multitalker babble at a 9 dB signal-to-noise ratio (S/N) (Cord et al, 1992). The SPRINT was developed by and used by the Army as a more valid predictor of communication abilities (than pure-tone thresholds or word-recognition in quiet) for issues involving fitness for duty from a hearing perspective of Army personnel. The Words-in-Noise test (WIN) is a slightly different word-recognition task in a fixed level multitalker babble with 10 NU-6 words presented at each of 7 S/N from 24 to 0 dB S/N in 4 dB decrements (Wilson, 2003; Wilson and McArdle, 2007). For the two instruments, both the babble and the speakers of the words are different. The SPRINT uses all 200 NU-6 words, whereas the WIN uses a maximum of 70 words. Purpose: The purpose was to compare recognition performances by 24 young listeners with normal hearing and 48 older listeners with sensorineural hearing on the SPRINT and WIN protocols. Research Design: A quasi-experimental, mixed model design was used. Study Sample: The 24 young listeners with normal hearing (19 to 29 years, mean = 23.3 years) were from the local university and had normal hearing (≤20 dB HL; American National Standards Institute, 2004) at the 250–8000 Hz octave intervals. The 48 older listeners with sensorineural hearing loss (60 to 82 years, mean = 69.9 years) had the following inclusion criteria: (1) a threshold at 500 Hz between 15 and 30 dB HL, (2) a threshold at 1000 Hz between 20 and 40 dB HL, (3) a three-frequency pure-tone average (500, 1000, and 2000 Hz) of ≤40 dB HL, (4) word-recognition scores in quiet ≥40%, and (5) no history of middle ear or retrocochlear pathology as determined by an audiologic evaluation. Data Collection and Analysis: The speech materials were presented bilaterally in the following order: (1) the SPRINT at 50 dB HL, (2) two half lists of NU-6 words in quiet at 60 dB HL and 80 dB HL, and (3) the two 35-word lists of the WIN materials with the multitalker babble fixed at 60 dB HL. Data collection occurred during a 40–60 minute session. Recognition performances on each stimulus word were analyzed. Results: The listeners with normal hearing obtained 92.5% correct on the SPRINT with a 50% point on the WIN of 2.7 dB S/N. The listeners with hearing loss obtained 65.3% correct on the SPRINT and a WIN 50% point at 12.0 dB S/N. The SPRINT and WIN were significantly correlated (r = −0.81, p < .01), indicating that the SPRINT had good concurrent validity. The high-frequency, pure-tone average (1000, 2000, 4000 Hz) had higher correlations with the SPRINT, WIN, and NU-6 in quiet than did the traditional three-frequency pure-tone average (500, 1000, 2000 Hz). Conclusions: Graphically and numerically the SPRINT and WIN were highly related, which is indicative of good concurrent validity of the SPRINT.

Download Full-text