In recent times, the use of industrial Heating, Ventilation, and Air Conditioning (HVAC) systems has had a substantial increase due to various social-economic reasons. Various studies are conducted to address the concerns of noise generated by HVAC systems in particular from compressors; noise pollution affects the environment we live in and should therefore be mitigated as much as possible. For this purpose, an active noise reduction system is introduced to help alleviate noise levels at both low and high frequency ranges of the HVAC system. This paper demonstrates an implementation of Active Noise Reduction (ANR) system using Least Mean Square (LMS) and Filtered-x Least Mean Square (FxLMS) algorithms, which have been tested based on simulated noise scenarios. A comparison between both algorithms will be performed based on simulation of different frequency ranges which closely resembles the operating scenario of the industrial HVAC systems. Also, different values of step-size affecting the processing of the ANR system during the convergence rate and stability state will also be investigated and discussed.

Speech synthesis plays a pivotal role nowadays. It can be found in various daily applications such as in mobile phones, navigation systems, languages learning software and so on. In this study, a Malay language speech synthesizer was designed using hidden Markov model to improve the performance of current Malay speech synthesizer and also extend Malay speech technology. Statistical parametric method was utilized in this study. The database was constructed to be balanced with all the phonetic sample appeared in Malay language. The results were rated by 48 listeners and obtained a moderate high rating ranging from 3.79 to 4.23 out of 5. The computed Word Error Rate is 7.1%. The total file size is less than 2 Megabytes which means it is suitable to be embedded into daily application. In conclusion, a Malay language speech synthesizer was designed using statistical parametric method with hidden Markov model. The output speech was verified to be good in quality. The file size is small indicates the feasibility to be used in embedded system.

The preparation of training data for statistical parametric speech synthesis can be sophisticated. To ensure the good quality of synthetic speech, high quality low noise recording must be prepared. The preparation of recording script can be also tremendous from words collection, words selection and sentences design. It requires tremendous human effort and takes a lot of time. In this study, we used alternative free source of recording and text such as audio-book, clean speech and so on as the training data. Some of the free source can provide high quality recording with low noise which is suitable to become training data. Statistical parametric speech synthesis method applying Hidden Markov Model (HMM) has been used. To test the reliability of synthetic speech, perceptual test has been conducted. The result of naturalness test is fairly reasonable. The intelligibility test showed encouraging result. The Word Error Rate (WER) for normal synthetic sentences is below 15% while for Semantically Unpredictable Sentences (SUS) is averagely in 30%. In short, using free and ready source as training data can leverage the process of preparing training data while obtaining motivating synthetic result.

Using found data as training data in statistical parametric speech synthesis can alleviate various problems in tedious database construction. However, the extra silences resided in found data degrades the quality of synthetic speech. Therefore, in this study, silence cutter was created to eliminate the extra silences in the training data. The motivation is the extra silences would be incorrectly assigned to training script and result in unnatural synthetic speech. Therefore, in this study, a Malay speech synthesis system has been constructed using found data from internet. Silence cutter has been utilized to cut out extra silences. The synthetic speech using found data with and without silence cutter was verified and compared to find out the effect of silence cutter. Result showed that silence cutter has help to improve synthetic speech naturalness and reduce the Word Error Rate (WER) in intelligibility test. In short, using found data can alleviate the problem of preparing high quality training data and silence cutter can be used to refine the found data to generate better quality of synthetic speech.

Abstract:In this study, we aim to reduce the human effort in preparing training data for synthesizing human speech and improve the quality of synthetic speech. In spite of the learning-from-data used to train the statistical models, the construction of a statistical parametric speech synthesizer involves substantial human effort, especially when using imperfect data or working on a new language. Here, we use lightly-supervised methods for preparing the data and constructing the text-processing front end. This initial system is then iteratively improved using active learning in which feedback from users is used to disambiguate the pronunciation system in our chosen language, Malay. The data are prepared using speaker diarisation and lightly-supervised text-speech alignment. In the front end, graphemebased units are used. The active learning used small amounts of feedback from a listener to train a classifier. We report evaluations of two systems built from high-quality studio data and lower-quality `found' data respectively and show that the intelligibility of each can be improved using active learning.