China 
Africa 
Explore chapters and articles related to this topic
Bottom Up Speech Recognition
Published in Robert H. Chen, Chelsea Chen, Artificial Intelligence, 2022
However, context-dependent models clearly require a great deal of speech data to be accurate, so for more efficient use of the data, the HMM is divided into sub-HMMs for each triphone (sequence of three phonemes) and HMM decision-trees clustered by alpha-beta pruning to associate different speech states.
Refinement of HMM Model Parameters for Punjabi Automatic Speech Recognition (PASR) System
Published in IETE Journal of Research, 2018
Virender Kadyan, Archana Mantri, R. K. Aggarwal
In the Punjabi language, till date, only a few types of researches are being carried out in speech analysis part. It is still considered as a less resource language due to unavailability of resources in it. Due to dialectal and tonal characteristics of Punjabi, a large corpus is required to build a speech to text. Today, a number of researches have been carried on text analysis aspect [18,19] or only a few researches have been carried out by researchers on the building of an ASR system in the Punjabi language like Kumar [20] worked on speaker dependent small vocabulary isolated words using DTW technique. Dua et al. [16] presented a speaker dependent and speaker independent 115 isolated words Punjabi ASR system from eight speakers using HMM technique for an acoustic model generation. They further extended their research work using connected words of these 115 words using HMM at the back end and MFCC at the front end [21]. Ghai and Singh [22] proposed a continuous speech recognition system using 100 sentences from nine speakers using triphone modeling. Lata and Arora [15] studied the impact of /h/ sound using Praat and MATLAB software on malwa database. Mittal [23] presented three modes of data, i.e. read, lecture, and conversation speech collected mainly from Malwai region and it was tested using continuous density HMM with the help of HTK toolkit. Singh et al. [24] have discussed five Punjabi tonemes based on their position and performed analysis on 150 unique words collected from 10 speakers.