Automatic Speech Segmentation Research Articles

Amharic Speech Search Using Text Word Query Based on Automatic Sentence-like Segmentation

More than 7000 languages are spoken in the world today. Amharic is one of the languages spoken in the East African country Ethiopia. A lot of speech data is being made every day in different languages as machines are getting better at processing and have improved storing capacity. However, searching for a particular word with its respective time frame inside a given audio file is a challenge. Since Amharic has its own distinguishing characteristics, such as glottal, palatal, and labialized consonants, it is not possible to directly use models that are developed for other languages. A popular approach in developing systems for searching particular information in speech involves using an automatic speech recognition (ASR) module that generates the text version of the speech where the word or phrase is searched based on text query. However, it is not possible to transcribe a long audio file without segmentation, which in turn affects the performance of the ASR module. In this paper, we are reporting our investigation on the effects of manual and automatic speech segmentation of Amharic audio files in a spiritual domain. We have used manual segmentation as a baseline for our investigation and found out that sentence-like automatic segmentation resulted in a word error rate (WER) closer to the WER achieved on the manually segmented test speech. Based on the experimental results, we propose Amharic speech search using text word query (ASSTWQ) based on automatic sentence-like segmentation. Since we have achieved lower WER using the previously developed speech corpus, which is in a broadcast news domain, together with the in-domain speech corpus, we recommend using both in- and out-domain speech corpora to develop the Amharic ASR module. The performance of the proposed ASR is a WER of 53% that needs further improvement. Combining two language models (LMs) developed using training text from the two different domains (spiritual and broadcast news) allowed a WER reduction from 53% to 46%. Therefore, we have developed two ASSTWQ systems using the two ASR modules with WERs of 53% and 46%.

Russian Sign Language Database for Clinical Use: Data and Annotation Peculiarities

This work has been carried out by the members of the Laboratory of Speech and Multimodal Interfaces of the St. Petersburg Federal Research Center of the Russian Academy of Sciences within an interdisciplinary research project aimed at the creation of an automatic Russian sign language translation system. The paper presents the design of a Russian sign language digital database for a specific subject area, namely, «The first visit to doctor».Our database is meant to be used first of all a dataset for training neural network-based systems of automatic translation from Russian sign language. But also, it can be of interest for linguistics of sign languages in general since the approach elucidates solution of a number of distorting phenomena typical for continuous sign language tracking, such as epenthesis, assimilation, reduction, and hold deletion.The principal difference between the presented video data and other datasets developed for similar purposes is the use of continuous sign utterances and elements of Russian sign language proper instead of the so-called “signed Russian” (that is, the visual form of the Russian spoken language), popular in deaf schooling.One of the most challenging problems, dealt with in the paper, is an automatic segmentation of sign speech into separate meaningful units with clearly defined boundaries. Its solution is hampered by signs’ deformations in continuous speech. In this paper, the segmentation is carried out within the framework of the movement-hold model. This approach allows extraction of their functional core as well as annotation of possible changes likely to appear in different signs’ segments. Accordingly, each utterance was subdivided into segments of motion and hold, and the resulting sign schemes were then entered into a separate directory containing information about the main parameters of each hold, including the possibilities of change due to immediate environment. The result of this work is a full decomposition of the signs, forming the database which can find its application in different statistical models of Russian Sign Language.Among other linguistic peculiarities of the database should be noted lexical variability of signs, mouthing and code switching between Russian sign language and “signed Russian”. Also, our signers learned Russian sign language in different regions of Russia, thus the collected data is potentially a source for research in dialectal variability of Russian sign language.

An Efficient Syllable-Based Speech Segmentation Model Using Fuzzy and Threshold-Based Boundary Detection

To develop a high-quality TTS system, an appropriate segmentation of continuous speech into the syllabic units plays a vital role. The significant objective of this research work involves the implementation of an automatic syllable-based speech segmentation technique for continuous speech of the Hindi language. Here, the parameters involved in the segmentation process are optimized to segment the speech syllables. In addition to this, the proposed iterative splitting process containing the optimum parameters minimizes the deletion errors. Thus, the optimized iterative incorporation can discard more insertions without merging the frequent non-iterative incorporation. The mixture of optimized iterative and iterative incorporation provides the best accuracy with the least insertion and deletion errors. The segmentation output based on different text signals for the proposed approach and other techniques namely GA, PSO and SOM is accurately segmented. The average accuracy obtained for the proposed approach is high with 97.5% than GA, PSO and SOM. The performance of the proposed algorithm is also analyzed and gives better-segmented accuracy when compared with other state-of-the-art methods. Here, the syllable-based segmented database is suitable for the speech technology system for Hindi in the travel domain.

An automatic speech segmentation tool based on multiple acoustic parameters

Speech segmentation is required not only for linguistic research based on oral corpora, but had become essential for natural language processing. Many researchers have developed different approaches to deal with the need of automatic segmentation of speech data. In this paper we discuss some of the prosodic parameters used as cues for boundary identification and present an ongoing project for the automatic segmentation of spontaneous speech developed for Brazilian Portuguese.

A REVIEW: METHODS OF AUTOMATIC SPEECH SEGMENTATION

Segmentation is a process of dividing a speech signal into the basic units of language. Segmentation of the speech signals is one of the most important tasks in automatic speech processing systems. This paper proposes a review of methods of automatic speech segmentation. Moreover, methods of wavelet and Hilbert- Huang transformations and techniques based on hidden Markov models are considered.

Automated Waves Files Splitting

The ASS (Automatic Speech Segmentation) Technique is used in this article to segment spontaneous speech into syllable-like units. The segmentation of the acoustic signal into syllabic units is an essential step in the construction of a syllable-centric ASS system. The purpose of this article is to determine the smallest unit of speech that should be regarded when writing. Any voice recognition system may be trained. In a few Indian cities, technologies for continuous voice recognition have been created. Hindi and Tamil are examples of such languages. This article examines the statistical characteristics of Punjabi syllables and how they may be used to reduce the number of syllables in sentence. During voice recognition, the search area is expanded. We explain how to perform the majority of the segmentation in this article automatically. The frequency of syllables and the number of syllables in each word are shown. We suggest the following: For objective evaluation of stuttering disfluencies, an automated segmentation technique for syllable repetition in read speech was developed. It employs a novel method and consists of three stages: feature extraction, rule matching, and segmentation.

Significance of spectral cues in automatic speech segmentation for Indian language speech synthesizers

Building speech synthesis systems for Indian languages is challenging owing to the fact that digital resources for these languages are hardly available. Vocabulary independent speech synthesis requires that a given text is split at the level of the smallest sound unit, namely, phone. The waveforms or models of phones are concatenated to produce speech. The waveforms corresponding to that of the phones are obtained manual (listening and marking) when digital resources are scarce. But the manual labeling of speech data (also known as speech segmentation) can lead to inconsistencies as the duration of phones can be as short as 10ms.The most common approach to automatic segmentation of speech is to perform forced alignment using monophone hidden Markov models (HMMs) that have been obtained using embedded re-estimation after flat start initialization. These results are then used in neural network frameworks to build better acoustic models for speech synthesis/recognition. Segmentation using this approach requires large amounts of data and does not work very well for low resource languages. To address the issue of paucity of data, signal processing cues like short-term energy (STE) and sub-band spectral flux (SBSF) are used in tandem with HMM based forced alignment for automatic speech segmentation.STE and SBSF are computed on the speech waveforms. STE yields syllable boundaries, while SBSF provides locations of significant change in spectral flux that are indicative of fricatives, affricates, and nasals. STE and SBSF cannot be used directly to segment an utterance. Minimum phase group delay based smoothing is performed to preserve these landmarks, while at the same time reducing the local fluctuations. The boundaries obtained with HMMs are corrected at the syllable level, wherever it is known that the syllable boundaries are correct. Embedded re-estimation of monophone HMM models is again performed using the corrected alignment. Thus, using signal processing cues and HMM re-estimation in tandem, robust monophone HMM models are built. These models are then used in Gaussian mixture model (GMM), deep neural network (DNN) and convolutional neural network (CNN) frameworks to obtain state-level frame posteriors. The boundaries are again iteratively corrected and re-estimated.Text-to-speech (TTS) systems are built for different Indian languages using phone alignments obtained with and without the use of signal processing based boundary corrections. Unit selection based and statistical parametric based TTS systems are built. The result of the listening tests showed a significant improvement in the quality of synthesis with the use of signal processing based boundary correction.

A Method for Automatic Vietnamese Speech Segmentation

In speech synthesis and recognition, the segmentation is an important step. The result of further steps depend completely on this process. There are several effective segmentation method in the literature, but for Vietnamese speech, researchers usually base on their experience to set the length while using sliding window. It causes an inefficient segmentation; and they need to try with the other value (length of voice). In this paper, we propose a method supporting in segmentation for Vietnamese speech and automatically determine the suitable length of voices and silent pause. We firstly estimate, by experimenting, the min and average length of a voice and a silent pause for Vietnamese speech in three main type speaking (slow, normal and fast). Then, based on these values, we start to segment the voice and pause by sliding window with proposed algorithm. Experiment results show that the proposed method can be used to effectively segment the Vietnamese speech.

Unsupervised Phonetic Segmentation of Classical Arabic Speech Using Forward and Inverse Characteristics of the Vocal Tract

Automatic segmentation of speech is about identifying boundaries of phonemes in a given utterance. This paper presents a strategy driven by cosine distance similarity scores for identifying phoneme boundaries. The proposed strategy helps in the selection of appropriate feature extraction technique for speech segmentation applications. After assessing various state-of-the-art speech processing techniques, a novel combination of forward and inverse characteristics of vocal tract (FICV) is developed. The proposed technique is evaluated on Classical Arabic dataset. Extensive experiments are made to compare the proposed technique with state-of-the-art techniques, including the hidden Markov model-based forced alignment procedures. The results show that proposed technique has total error rate of 14.48%, while the accuracy is 85.2% within 10 ms alignment error. When compared with the existing state-of-the-art technique, the proposed technique outperforms by 12.29% and 22.73% in terms of error rates and alignment accuracies, respectively, which signifies the potential of using FICV in speech segmentation.

Optimizing Parameters of Automatic Speech Segmentation into Syllable Units

Optimizing Parameters of Automatic Speech Segmentation into Syllable Units

민주당 대의원제도 개혁사에서 본 2018년 수퍼대의원 제도의 개혁 - 2016년∼2018년 개혁의 내용, 원인, 그리고 함의

Spoken language translation (SLT) has become very important in an increasingly globalized world. Machine translation (MT) for automatic speech recognition (ASR) systems is a major challenge of great interest. This research investigates that automatic sentence segmentation of speech that is important for enriching speech recognition output and for aiding downstream language processing. This article focuses on the automatic sentence segmentation of speech and improving MT results. We explore the problem of identifying sentence boundaries in the transcriptions produced by automatic speech recognition systems in the Polish language. We also experiment with reverse normalization of the recognized speech samples.

Syllable Segmentation of Tamil Speech Signals Using Vowel Onset Point and Spectral Transition Measure

Segmentation plays vital role in speech recognition systems. An automatic segmentation of Tamil speech into syllable has been carried out using Vowel Onset Point (VOP) and Spectral Transition Measure (STM). VOP is a phonetic event used to identify the beginning point of the vowel in speech signals. Spectral Transition Measure is performed to find the significant spectral changes in speech utterances. The performance of the proposed syllable segmentation method is measured corresponding to manual segmentation and compared with the exiting syllable method using VOP and Vowel Offset Point (VOF). The result of the experiments shows the effectiveness of the proposed system.

Automatic syllable segmentation algorithm of Chinese speech based on MF-DFA

Automatic speech segmentation algorithm plays an important role in speech recognition and spoken term detection. A method called automatic syllable segmentation of Chinese speech based on multi-fractal detrended fluctuation analysis (MF-DFA) is explored in this study. The algorithm attempts to improve the precision and robustness of Chinese syllable segmentation. Firstly, the multi-fractal characteristics of Chinese syllables based on MF-DFA are explored. Secondly, to solve the problem with the unclear boundary of adjacent finals, which leads to the unsatisfactory precision rate of Chinese syllable segmentation in existing algorithms, two-stage voiced decision algorithm is introduced. Finally, the generation of dividing point works by detecting the extreme points of the first-order differential curve for each voiced segment. The experimental results indicated that the multi fractal characteristics based on MF-DFA possess good distinction and robustness, and the proposed algorithm outperforms the earlier approaches in terms of the performance of Chinese syllable segmentation even in low SNR.

Multilingual processing of speech via web services

A new software paradigm `Software as a Service' based on web services is proposed for multilingual linguistic tools and exemplified with the BAS CLARIN web services. Instead of traditional tool development and distribution the tool functionality is implemented on a highly available server that users or applications access via HTTP requests. As examples we describe in detail five multilingual web services for speech science operational since 2012 and discuss the benefits and drawbacks of the new paradigm as well as our experiences with user acceptance and implementation problems. The services include automatic segmentation of speech, grapheme-to-phoneme conversion, syllabification, speech synthesis, and optimal symbol sequence alignment.

Automatic speech segmentation using throat-acoustic correlation coefficients

Abstract This work considers one of the approaches to the solution of the task of discrete speech signal automatic segmentation. The aim of this work is to construct such an algorithm which should meet the following requirements: segmentation of a signal into acoustically homogeneous segments, high accuracy and segmentation speed, unambiguity and reproducibility of segmentation results, lack of necessity of preliminary training with the use of a special set consisting of manually segmented signals. Development of the algorithm which corresponds to the given requirements was conditioned by the necessity of formation of automatically segmented speech databases that have a large volume. One of the new approaches to the solution of this task is viewed in this article. For this purpose we use the new type of informative features named TAC-coefficients (Throat-Acoustic Correlation coefficients) which provide sufficient segmentation accuracy and effi- ciency.

Proposta de delimitação de elementos discursivos baseada na prosódia

This article aims to show the consistent alignment between an automatic segmentation of speech in stress groups and the delimitation of discourse segments according to Grosz and Sidner model (1986). The comparison between the manual segmentation of spontaneous speech in discourse segments that carry specific purposes and the automatic segmentation of stress groups shows a persistent concentration of a specific number of stress groups in each discourse segment. Based on the previous findings (Lucente, 2012) this article compares a manual delimitation of discourse segments and an automatic delimitation of the same discourse segments according to the correlation between the number of stress groups in each discourse segment. The results point to a robust coincidence between the two segmentations.

Automatic Speech Segmentation Using Hybrid Wavelet Features and HMM

In this research, a novel feature set is used to automatically segment speech signal. Automatic segmentation is veryuseful especially for large database. A hybrid features model is created from wavelet packet analysis and mel-scale is used to train Hidden Markov Model (HMM) for phone boundary detection. HMM is implemented using the Hidden Markov Model Toolkit (HTK).The database (Ked-TIMIT) is used for result verifications and Mel Frequency Cepstral Coefficients (MFCC) is used as reference for evaluating the results of the proposed Hybrid model. The results are categorized for vowels, consonants and short phones. Phone duration and start location are used as metrics to evaluate the system success rate. Success rate of 74% is achieved for consonant detection, 72% for vowel detection and 58% for short phone detection. Using the simple metric that relies only on boundary locations but ignoring duration, the achieved results are 92.5% for consonant detection, 90% for vowel detection and 77.5% for short phoneme detection. In addition to boundary detection the proposed hybrid model is utilized to compare newly developed features called Mel scale Best Tree Encoding (Mel-BTE ) to the mostly used popular features MFCC along with all experiments using the same database. The relative results for Mel-BTE with respect to MFCC are 94.77% for consonant detection, 87.5% for vowel detection and 93.33% for short phoneme detection.

AAYUDHA: A Tool for Automatic Segmentation and Labelling of Continuous Tamil Speech

An effective way of communication between human is now becoming an alternative way to communicate between human and machine. This alternative way is now-a-days used in many real time systems for faster, easier and comfortable response and communication. Speech segmentation and labelling are the process that lay as a key to decide the accuracy of several speech related research. A tool AAYUDHA is proposed that enables automatic segmentation and labelling of continuous speech in Tamil. Two different segmentation algorithms, one based on Fast Fourier Transform (FFT) feature set and 2D filtering and other based on Discrete Wavelet Transform (DWT) feature set and its energy variation in different sub-bands are implemented. The segmentation accuracy of those algorithms is analyzed. Further the segmented speech is labelled using a baseline Hidden Markov Model (HMM) based acoustic model. A speech corpus named KAZHANGIYAM is created which includes the recorded Tamil speech of various speakers. The database also includes the information of manually segmented data of those speech data. This speech corpus is used to analyze the accuracy of the algorithms used in the proposed tool. This tool concentrates on the phonetic level segmentation of Tamil speech. The tool shows an acceptable segmentation and labelling accuracy.

Enhancements in Statistical Spoken Language Translation by De-normalization of ASR Results

Spoken language translation (SLT) has become very important in an increasingly globalized world. Machine translation (MT) for automatic speech recognition (ASR) systems is a major challenge of great interest. This research investigates that automatic sentence segmentation of speech that is important for enriching speech recognition output and for aiding downstream language processing. This article focuses on the automatic sentence segmentation of speech and improving MT results. We explore the problem of identifying sentence boundaries in the transcriptions produced by automatic speech recognition systems in the Polish language. We also experiment with reverse normalization of the recognized speech samples.

Automatic continuous speech recogniser for Dravidian languages using the auto associative neural network

In recent times with the extensive improvement of computers, numerous methods of data interchange between man and computer are revealed. It aims to provide an efficient way for human to communicate with computers exclusively for people with disabilities who face diversity of obstacles while using computers. This paper predominantly focuses on developing an efficient speech recognition system for Dravidian languages such as Tamil, Malayalam, Telugu and Kannada. The proposed CSR system comprises of four steps namely pre-processing, feature extraction, automatic continuous speech segmentation and classification. The most powerful and widely used short term energy and zero crossing rate is used for continuous speech segmentation and Mel frequency cepstral coefficients MFCC, linear predictive cepstral coefficients LPCC and shifted delta cepstrum SDC feature extractions are used for recognition system. Experiments are carried out with real time Dravidian languages speech signal. It is observed from the results that the proposed system gives significant results in AANN classifier with MFCC feature when compared with LPCC and SDC features.