Open Speech Recognition Research Articles

The third ‘CHiME’ speech separation and recognition challenge: Analysis and outcomes

This paper presents the design and outcomes of the CHiME-3 challenge, the first open speech recognition evaluation designed to target the increasingly relevant multichannel, mobile-device speech recognition scenario. The paper serves two purposes. First, it provides a definitive reference for the challenge, including full descriptions of the task design, data capture and baseline systems along with a description and evaluation of the 26 systems that were submitted. The best systems re-engineered every stage of the baseline resulting in reductions in word error rate from 33.4% to as low as 5.8%. By comparing across systems, techniques that are essential for strong performance are identified. Second, the paper considers the problem of drawing conclusions from evaluations that use speech directly recorded in noisy environments. The degree of challenge presented by the resulting material is hard to control and hard to fully characterise. We attempt to dissect the various ‘axes of difficulty’ by correlating various estimated signal properties with typical system performance on a per session and per utterance basis. We find strong evidence of a dependence on signal-to-noise ratio and channel quality. Systems are less sensitive to variations in the degree of speaker motion. The paper concludes by discussing the outcomes of CHiME-3 in relation to the design of future mobile speech recognition evaluations.

Hearing Performance in Noise of Cochlear Implant Patients versus Severely-Profoundly Hearing-Impaired Patients with Hearing Aids:Rendimiento auditivo en ambiente ruidoso en pacientes con implante coclear versus hipoacúsicos profundos con auxiliar auditivo

It is possible for most post-lingually deaf patients to attain significant open speech recognition following cochlear implantation. In contrast, many severely-profoundly sensorineural hearing-impaired patients receive no benefit from their hearing aids, especially in situations with background noise. The aim of the study was to evaluate the speech recognition ability in quiet and in noise of post-lingually deaf adults implanted with Combi 40/40+ cochlear implants versus severely-profoundly sensorineural hearing-impaired patients fitted with hearing aids. For this purpose, we tested two groups of patients: one that had received cochlear implants (n=22) and a group of subjects with severe-profound sensorineural hearing impairment, fitted with hearing aids (n = 15). All of the patients were tested using the Hochmaier, Schultz, and Moser Discrimination Test in quiet and noise. The results of the study demonstrate that most of our cochlear implant patients received a substantial benefit from their implant, achieving scores of 70 to 100 per cent (mean, 90 per cent) for the numbers test and 10 to 72 per cent (mean, 43 per cent) for the monosyllable test 1 year after implantation. Even in situations with background noise, scores of 1 to 99 per cent (mean, 45.65 per cent) for a signal to noise ratio (SNR) of +15 dB 1 year following the implantation improved to 7 to 95 per cent (mean, 50.7 per cent) at 2 years and 8 to 99 per cent (mean, 60 per cent) at 3 years after implantation. These results are significantly (p<0.04) superior to the hearing aid patients' scores of 1 to 64.2 per cent (mean, 26.7 per cent) for a SNR of 15 dB. The results of the present study may have clinical implications in regard to selection of candidates for cochlear implantation.

Speech Perception Performance of Children with a Cochlear Implant Compared to that of Children with Conventional Hearing Aids: 1. The “Equivalent Hearing Loss” Concept

A new measure has been developed to quantify the speech perception performance of children with a cochlear implant (CI). The method summarizes the speech perception scores obtained on a battery of tests that ranges from very basal tasks up to open speech recognition. The overall performance of a child with a CI on the test battery at a certain time during follow-up is matched to that of a reference group of severely and profoundly hearing-impaired children with conventional hearing aids. This matching procedure results in the expression of the speech perception scores of a child with a CI as an “equivalent hearing loss” value. The equivalent hearing loss concept deals adequately with floor and ceiling effects which inevitably occur when a battery with such a large range of tests is used. To illustrate this, application of the procedure to three children with a CI showed that before implantation, while they were using conventional hearing aids, the equivalent hearing loss was above 120 dB hearing level (HL). At 3 years' follow-up the equivalent hearing loss improved to 70 dB HL in the two children with an aetiology of meningitis. This means that these children were performing as well as children in the reference group with a hearing loss of 70 dB HL. The child with congenital deafness showed minor improvements over time.

The restoration of hearing in neurofibromatosis type 2

Central electrical stimulation of the auditory pathway can allow hearing in patients suffering from deafness localized in the auditory nerve. Developments in a multi-channel auditory brainstem implant based on the Nucleus Mini 22 Cochlear implant with transcutaneous signal transmission is discussed. The devices have been implanted in nine European patients suffering from Neurofibromatosis Type 2. Preliminary speech perception results and patient satisfaction are encouraging, and the data presented include some limited open speech recognition.

Auditory temporal resolution and open speech recognition in cochlear implant recipients.

The recognition of speech events depends, among other factors, on the resolution function. The present study was conducted to evaluate the auditory temporal resolution function in cochlear implant (CI) subjects by a gap detection threshold (GDT) test. Fourteen postlingual deaf subjects who used the Nucleus 22 Channel CI system participated in the study. They were divided into two subgroups: (1) six patients who did not achieve a significant open speech recognition score; and (2) eight patients who obtained different levels of open speech recognition ability (OSRA). Gap detection thresholds were found for six noise burst durations: 85, 65, 52, 36, 25 and 10 msec. Absolute values of GDT in the CI group were longer compared with those obtained from normal-hearing subjects. The CI values varied in consistence with burst duration; namely, as burst duration decreased, GD increased. This trend was found in CI recipients with OSRA only. Examination of the relationship between GDT and OSRA indicated that the majority of CI recipients who had OSRA showed lower GDT.