Hmm

An offline cursive handwriting recognition system, based on hybrid of Neural Networks (NN) and Hidden Markov Models (HMM), is described in this paper. Applying SegRec principle, the recognizer does not make hard decision at the character segmentation process. Instead, it delays the character segmentation to the recognition stage by generating a segmentation graph that describes all possible ways to segment a word into letters. To recognize a word, the NN computes the observation probabilities for each segmentation candidates (SCs) in the segmentation graph. Then, using concatenated letter-HMMs, a likelihood is computed for each word in the lexicon by multiplying the probabilities over the best paths through the graph. We present in detail two approaches to train the word recognizer: 1). character-level training 2). word-level training. The recognition performances of the two systems are discussed. I.

The basic theory of Markov chains has been known to mathematicians and engineers for close to 80 years, but it is only in the past decade that it has been applied explicitly to problems in speech processing. One of the major reasons why speech models, based on Markov chains, have not been developed until recently was the lack of a method for optimizing the parameters of the Markov model to match observed signal patterns. Such a method was proposed in the late 1960's and was immediately applied to speech processing in several research institutions. Continued refinements in the theory and implementation of Markov modelling techniques have greatly enhanced the method, leading to a wide range of applications of these models. It is the purpose of this tutorial paper to give an introduction to the theory of Markov models, and to illustrate how they have been applied to problems in speech recognition.

An audio is a sound within audible range available to human. This range is study of all mechanical waves in gases, liquids and solids including vibration, and sound. Audio segmentation is most important preprocessing step in several audio processing applications with a significant impact e.g. on speech recognition system. This paper introduces a framework which analyzes the advantages of different segmentation and classification methods. The different audio segmentation methods like Bayesian Information Criterion (BIC), two phase audio segmentation method, metric-based method are studied. Audio segmentation plays important role in audio classification. After segmenting audio it is classified into different classes. The classification model consists of rule-based and model based.

Devanagari is the most popular script in India use by more than 500 million people. The handwriting recognition area has been researched extensively till date, whereas offline recognition of Devanagari script is progressing area of research. This system involves three steps, pre-processing, feature extraction and classification respectively. In the first stage words from input images are normalized. Then, a set of intensity features are extracted from each of the segmented words. Gradient features are extracted from an image Then structure-like features are also extracted. Finally, these features are applied in a combined scheme for classification. Intensity features & Gradient features are used to train a HMM classifier, whose results are re-ranked using structural features for improved recognition rate.

Abstract- This paper presents automatic speech recognition for continuous speech in Myanmar Language. Actually, a computer or a machine is not expected to understand what is uttered. But it is expected to be controlled via speech or to transcript the acoustic signal to symbols. This system will also address the issue of automatic word/sentence boundary detection in both quiet and noisy environments. Combinations of LPC, MFCC and GTCC techniques are used in feature extraction. MFCC features give the good discrimination of speech signal. LPC provides an accurate estimate of the speech parameters and it is also an efficient computational model of speech. DTW is used in the feature clustering and HMM is used in the recognition process. The HMM method is extended by combining it with the DTW algorithm in order to combine the advantages of these two powerful pattern recognition technique.

Globalization and growth of Internet users truly demands for almost all internet based applications to support local languages. Support of local languages can be given in all internet based applications by means of Machine Transliteration and Machine Translation. This paper provides the thorough survey on machine transliteration models and machine learning approaches used for machine transliteration over the period of more than two decades for internationally used languages as well as Indian languages. Survey
shows that linguistic approach provides better results for the closely related languages and probability based statistical approaches are good when one of the languages is phonetic and other is nonphonetic.Better accuracy can be achieved only by using Hybrid and Combined models.

This paper describes the development of an Arabic broadcast news transcription system. The presented system is a speaker-independent large vocabulary natural Arabic speech recognition system, and it is intended to be a test bed for further research into the open ended problem of achieving natural language man-machine conversation. The system addresses a number of challenging issues pertaining to the Arabic language, e.g. generation of fully vocalized transcription, and rule-based spelling dictionary. The developed Arabic speech recognition system is based on the Carnegie Mellon university Sphinx tools. The Cambridge HTK tools were also utilized at various testing stages. The system was trained on 7.0 hours of a 7.5 hours of Arabic broadcast news corpus and tested on the remaining half an hour. The corpus was made to focus on economics and sport news. At this experimental stage, the Arabic news transcription system uses five-state HMM for triphone acoustic models, with 8 and 16 Gaussian mixture distributions. The state distributions were tied to about 1680 senons. The language model uses both bi-grams and tri-grams. The test set consisted of 400 utterances containing 3585 words. The Word Error Rate (WER) came initially to 10.14 percent. After extensive testing and tuning of the recognition parameters the WER was reduced to about 8.61% for non-vocalized text transcription.

Urdu script-based languages’ character recognition has some technical issues not existing in other languages and makes these languages more complicated. Segmentation-based character recognition approach for handwritten Urdu, both Nasta’liq and Nasakh script-based languages, incorporates number of overhead and very less accurate as compared to segmentation free. This paper presents a segmentation-free approach for recognition of online Urdu handwritten script using hybrid classifier, HMM and fuzzy logic. Trained data set consisting of HMMs for each stroke is further classified into 62 sub-patterns based on the primary stroke shape at the beginning and end using fuzzy rule. Fuzzy linguistic variables based on language structure are used to model features and provide suitable result for large variation in handwritten strokes. Twenty-six time variant structural and statistical features are extracted for the base strokes. The fuzzy classification into sub-patterns increases the efficiency and decreases the computational complexity due to reduction in data set size. The hybrid HMM–fuzzy technique is efficient for large and complex data set. It provided 87.6% and 74.1% for Nasta’liq and Nasakh, respectively, on 1800 ligatures.

Naturalistic affective expressions change at a rate much slower than the typical rate at which video or audio is recorded. This increases the probability that consecutive recorded instants of expressions represent the same affective content. In this paper, we exploit such a relationship to improve the recognition performance of continuous naturalistic affective expressions. Using datasets of naturalistic affective expressions (AVEC 2011 audio and video dataset, PAINFUL video dataset) continuously labeled over time and over different dimensions, we analyze the transitions between levels of those dimensions (e.g., transitions in pain intensity level). We use an information theory approach to show that the transitions occur very slowly and hence suggest modeling them as first-order Markov models. The dimension levels are considered to be the hidden states in the Hidden Markov Model (HMM) framework. Their discrete transition and emission matrices are trained by using the labels provided with the training set. The recognition problem is converted into a best path-finding problem to obtain the best hidden states sequence in HMMs. This is a key difference from previous use of HMMs as classifiers. Modeling of the transitions between dimension levels is integrated in a multistage approach, where the first level performs a mapping between the affective expression features and a soft decision value (e.g., an affective dimension level), and further classification stages are modeled as HMMs that refine that mapping by taking into account the temporal relationships between the output decision labels. The experimental results for each of the unimodal datasets show overall performance to be significantly above that of a standard classification system that does not take into account temporal relationships. In particular, the results on the AVEC 2011 audio dataset outperform all other systems presented at the international competition.

The paper discusses an Amharic speaker independent continuous speech recognizer based on an HMM/ANN hybrid approach. The model was constructed at a context dependent phone part sub-word level with the help of the CSLU Toolkit. A promising result of 74.28% word and 39.70% sentence recognition rate was achieved. These are the best figures reported so far for speech recognition for the Amharic language.

W artykule przedstawiono wyniki badań dotyczących automatycznej detekcji wybranych wad wymowy u dzieci za pomocą automatycznego rozpoznawania mowy. Detekcja wady wymowy może być przeprowadzana pod kątem diagnozy lub terapii logopedycznej. Jednym z często spotykanych typów wad wymowy jest substytucja polegająca na podstawianiu prawidłowego fonemu w słowie innym fonemem tego samego języka. W pracy rozważano automatyczną detekcję substytucji pod kątem zastosowań do terapii logopedycznej. W przypadku terapii zadanie automatycznego rozpoznawania mowy upraszcza się do rozpoznawania dwóch głosek: prawidłowej lub nieprawidłowej, ustalonych na podstawie wcześniejszej diagnozy. Jako materiał badawczy przyjęto następujące pary fonemów będące częściami wypowiedzi języka polskiego: {s, sz}, {si, sz}, {c, cz}, {ci, cz}, {dz, drz}, {dzi, drz}. Nagrania pochodziły od dzieci z wadami wymowy oraz od osób, które naśladowały określone wady wymowy. Proces rozpoznawania wady wymowy składał się z dwóch zasadniczych etapów: ekstrakcji cech z sygnału mowy oraz klasyfikacji. W przypadku ekstrakcji cech zbadano dwie metody: standardową metodę MFCC (ang. Mel-frequency cepstral coefficients) oraz stosunkowo niedawno wprowadzoną metodę HFCC (ang. Human-factor cepstral coefficients). Na etapie klasyfikacji przebadano skuteczność rozpoznawania wad wymowy za pomocą czterech metod. Pierwszą badaną metodą była metoda nieliniowej transformacji czasowej (ang. dynamic time warping – DTW). Standardowa metoda DTW jest oparta na modelach całych słów. W przypadku rozważanego problemu terapii substytucji słowa są rozróżnianie jedynie za pomocą jednego fonemu. W takiej sytuacji standardowa metoda DTW często zawodzi zwłaszcza, że fonemy rozróżniające dwa słowa są najczęściej podobne do siebie akustycznie. Ponadto segmenty poza obszarem rozróżniających słowa fonemów często podlegają różnym zniekształceniom lub zakłóceniom, co może dać w rezultacie większą niż zwykle odległość DTW pomiędzy słowami tej samej klasy. Zaproponowano modyfikację standardowej metody DTW polegającą na obliczaniu odległości DTW jedynie między fonemami będącymi składowymi danego słowa. W proponowanym rozwiązaniu zakłada się, że klasa rozpoznawanego słowa jest znana oraz, że słowo może być wypowiedziane prawidłowo lub nieprawidłowo zgodnie z postawioną wcześniej diagnozą. Oprócz metod opartych na nieliniowej transformacji czasowej zbadano również metodę klasyfikacji bazującą na niejawnych modelach Markowa (ang. hidden Markov models – HMM). Metodę HMM badano dla modeli całych słów oraz dla modeli fonemów. W przeprowadzonych badaniach zaobserwowano następujące tendencje: • metoda DTW rozpoznająca fonemy dała wyższe skuteczności niż metoda DTW rozpoznająca słowa • metoda HMM dla modeli fonemów dała lepsze rezultaty w porównaniu z metodą HMM dla całych słów • Skuteczności rozpoznawania w oparciu o cechy HFCC były wyższe w porównaniu ze standardowymi cechami MFCC • W porównaniu z klasyfikatorem DTW, metoda HMM dała nieznacznie gorsze wyniki, jednak problem ten wymaga dalszych badań. Opracowane metody rozpoznawania mogą znaleźć zastosowanie w diagnostyce i terapii wad wymowy u dzieci. W szczególności można z ich wykorzystaniem dokonywać wykrywania substytucji następujących par fonemów: sz-s, cz-c oraz drz-dz do celów terapii logopedycznej. Dla par fonemów sz-si, cz-ci, drz-dzi muszą zostać opracowane inne, bardziej skuteczne metody. Potencjalne przyszłe kierunki badawcze obejmują zastosowanie metody analizy składowych głównych oraz analizy dyskryminacyjnej zarówno w metodzie DTW jak i HMM. Planuje się również dalszą optymalizację parametrów metody HMM. Opisywane w pracy badania były sponsorowane z grantu MNiI nr 1 H01F 046 28.