Abstract
The automatic speech recognition (ASR) model usually requires a large amount of training data to provide better results compared with the ASR models trained with a small amount of training data. It is difficult to apply the ASR model to non-standard speech such as that of cochlear implant (CI) patients, owing to privacy concerns or difficulty of access. In this paper, an effective finetuning and augmentation ASR model is proposed. Experiments compare the character error rate (CER) after training the ASR model with the basic and the proposed method. The proposed method achieved a CER of 36.03% on the CI patient’s speech test dataset using only 2 h and 30 min of training data, which is a 62% improvement over the basic method.
Highlights
Various automatic speech recognition (ASR) models have been proposed in recent years, including the recurrent neural network transducer (RNN-T) [1], Listen, Attend and
Speech understanding can be restored through cochlear implants in people with severe hearing loss, especially sensorineural hearing loss
Because the cochlear implant (CI) patient’s speech is already distorted, it can be confirmed that the augmentation that modifies the raw audio is not effective
Summary
Various automatic speech recognition (ASR) models have been proposed in recent years, including the recurrent neural network transducer (RNN-T) [1], Listen, Attend and. ASR models are trained using standard speech data sets [4,5]. People with non-standard speech cannot use ASR models trained with a standard speech dataset. We experimented with an effective ASR method to increase the recognition rate of the non-standard CI patients’ speech. The biggest hindrance in learning a non-standard dataset is to find sufficient data to train an ASR model [8]. Adversarial training can be used to generate training data by transforming standard speech into nonstandard speech [9]. We used a data augmentation technique and selected the augmentation method [10,11] used for standard speech.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
