Hasil Pencarian  ::  Simpan CSV :: Kembali

Hasil Pencarian

Ditemukan 8681 dokumen yang sesuai dengan query
cover
Mary, Leena
"This updated book expands upon prosody for recognition applications of speech processing. It includes importance of prosody for speech processing applications; builds on why prosody needs to be incorporated in speech processing applications; and presents methods for extraction and representation of prosody for applications such as speaker recognition, language recognition and speech recognition. The updated book also includes information on the significance of prosody for emotion recognition and various prosody-based approaches for automatic emotion recognition from speech."
Switzerland: Springer Cham, 2019
e20502221
eBooks  Universitas Indonesia Library
cover
Mary, Leena
"Extraction and representation of prosodic features for speech processing applications deals with prosody from speech processing point of view with topics including, the significance of prosody for speech processing applications, why prosody need to be incorporated in speech processing applications, and different methods for extraction and representation of prosody for applications such as speech synthesis, speaker recognition, language recognition and speech recognition."
New York: Springer, 2012
e20418411
eBooks  Universitas Indonesia Library
cover
New York: IEEE Press, c1979
621.381 9 AUT
Buku Teks  Universitas Indonesia Library
cover
Kondo, Kazuhiro
"This practical hands-on book shows speech intelligibility measurement methods so that the readers can start measuring or estimating speech intelligibility of their own system. The book also introduces subjective and objective speech quality measures, and describes in detail speech intelligibility measurement methods. It introduces a diagnostic rhyme test which uses rhyming word-pairs, and includes : an investigation into the effect of word familiarity on speech intelligibility. Speech intelligibility measurement of localized speech in virtual 3-D acoustic space using the rhyme test. Estimation of speech intelligibility using objective measures, including the ITU standard PESQ measures, and automatic speech recognizers."
Berlin: [, Springer], 2012
e20398888
eBooks  Universitas Indonesia Library
cover
Coleman, John R.
New York: Cambridge University Press, 2005
410.285 COL i
Buku Teks  Universitas Indonesia Library
cover
Mohammad Salman Alfarisi
"

Salah satu permasalahan yang terdapat pada sistem Automatic Speech Recognition (ASR) yang sudah ada adalah kurangnya transparansi dalam penanganan data suara, yang tentunya membuat adanya keraguan terhadap privasi data tersebut. Di sisi lainnya, untuk mengembangkan sebuah sistem ASR yang memiliki akurasi memadai dan dapat bekerja secara luring membutuhkan jumlah data yang banyak, khususnya data suara yang sudah diiringi dengan transkripnya. Hal ini menjadi salah satu hambatan utama pengembangan sistem pengenalan suara, terutama pada yang memiliki sumber daya minim seperti Bahasa Indonesia. Oleh karena itu, dalam penelitian ini dilakukan perancangan sistem pengenalan suara otomatis berbasis model wav2vec 2.0, sebuah model kecerdasan buatan yang dapat mengenal sinyal suara dan mengubahnya menjadi teks dengan akurasi yang baik, meskipun hanya dilatih data dengan label yang berjumlah sedikit. Dari pengujian yang dilakukan dengan dataset Common Voice 8.0, model wav2vec 2.0 menghasilkan WER sebesar 25,96%, dua kali lebih baik dibandingkan dengan model Bidirectional LSTM biasa yang menghasilkan 50% namun membutuhkan jumlah data dengan label 5 kali lipat lebih banyak dalam proses pelatihan. Namun, model wav2vec membutuhkan sumber daya komputasi menggunakan 2 kali lebih banyak RAM dan 10 kali lebih banyak memori dibandingkan model LSTM


One of the main problems that have plagued ready-to-use Automatic Speech Recognition (ASR) Systems is that there is less transparency in handling the user’s voice data, that has raised concerns regarding the privacy of said data. On the other hand, developing an ASR system from scratch with good accuracy and can work offline requires a large amount of data, more specifically labeled voice data that has been transcribed. This becomes one of the main obstacles in speech recognition system development, especially in low-resourced languages where there is minimal data, such as Bahasa Indonesia. Based on that fact, this research conducts development of an automatic speech recognition system that is based on wav2vec 2.0, an Artificial Model that is known to recognize speech signals and convert it to text with great accuracy, even though it has only been trained with small amounts of labeled data. From the testing that was done using the Common Voice 8.0 dataset, the wav2vec 2.0 model produced a WER of 25,96%, which is twice as low as a traditional Bidirectional LSTM model that gave 50% WER, but required 5 times more labeled data in the training process. However, the wav2vec model requires more computational resource, which are 2 times more RAM and 10 times more storage than the LSTM model.

"
Depok: Fakultas Teknik Universitas Indonesia, 2022
S-Pdf
UI - Skripsi Membership  Universitas Indonesia Library
cover
Arief Saferman
"

Selama masa pandemi COVID-19, teknologi Automatic Speech Recognition (ASR) menjadi salah satu fitur yang sering digunakan pada komputer untuk mencatat di kelas online secara realtime. Teknologi ini akan bekerja dimana setiap suara yang muncul akan langsung dikenali dan dicatat pada halaman terminal. Dalam penelitian ini, model ASR Wav2Letter akan digunakan menggunakan CNN (Convolution Neural Network) dengan loss function CTC (Connectionist Temporal Classification) dan ASG (Auto Segmentation Criterion). Selama proses pembuatannya, berbagai hyperparameter acoustic model dan language model dari model ASR Wav2Letter terkait dengan implementasi batch normalization¸ learning-rate, window type, window size, n-gram language model, dan konten language model diuji pengaruh variasinya terhadap performa model Wav2Letter. Dari pengujian tersebut, ditemukan bahwa model ASR Wav2Letter menunjukkan performa paling baik ketika acoustic model menggunakan metode ASG dengan learning-rate 9 × 10−5 , window size 0.1, window type Blackman, serta 6-gram language model. Berdasarkan hasil akurasi WER CTC unggul 1,2% dengan 40,36% berbanding 42,11% dibandingkan ASG, namun jika dilihat lamanya epoch dan ukuran file model, loss function ASG memiliki keunggulan hampir dua kalinya CTC, dimana ASG hanya membutuhkan setengah dari jumlah epoch yang dibutuhkan oleh CTC yakni 24 epoch berbanding dengan 12 epoch dan ukuran file model ASG setengah lebih kecil dibandingkan CTC yakni 855,2 MB berbanding dengan 427,8 MB. Pada pengujian terakhir, model ASR Wav2Letter dengan loss function ASG mendapatkan hasil terbaik dengan nilai WER 29,30%. Berdasarkan hasil tersebut, model ASR Wav2Letter dengan loss function ASG menunjukkan perfoma yang lebih baik dibandingkan dengan CTC.


During the COVID-19 pandemic, Automatic Speech Recognition technology (ASR) became one of features that most widely used in computer to note down online class in real-time. This technology works by writing down every word in terminal from voice that is recognized by the system. ASR Wav2Letter model will use CNN (Convolutional Neural Network) with loss function CTC (Connectionist Temporal Classification) and ASG (Auto Segmentation Criterion). While developing Wav2Letter, various hyperparameter from acoustic model and language model is implemented such as batch normalization, learning rate, window type, window size, n-gram language model, and the content of language model are examined against the performance of Wav2Letter model. Based on those examination, Wav2Letter shows best performance when it uses ASG loss function learning rate 9 × 10−5 , window size 0.1, window type Blackman, and 6-gram language model. With that configuration, WER of CTC outplay ASG around 1.2% with 40.36% compare to 42,11%, but another parameter shows ASG are way more superior than CTC with less time epoch training which are 24 epoch for CTC against 12 epoch for ASG and the size of memory model shows CTC has bigger size than ASG with 855.2 MB against 427.8 MB. In the last test, ASR Wav2Letter model with ASG loss function get the best WER value around 29.3%. Based on those results, ASR Wav2Letter Model shows its best performance with ASG loss function than CTC.

"
Depok: Fakultas Teknik Universitas Indonesia, 2022
S-Pdf
UI - Skripsi Membership  Universitas Indonesia Library
cover
"In the current study, two experiments are reported that investigated the effects of simple white noise and mixture of white noise and other sounds on perception of speech. In both experiments, university students were recruited to listen to short sentences under various sound masking conditions. Experiment 1, where standard sets of speakers were used for both speech and masking stimuli, has shown that, compared to baseline where there was no masking sound, the participants had significantly greater difficulties in understanding the sentences where the average level of understanding was 28% for the white noise condition and 20% for the mixed noise condition in which white noise was mixed with pink noise and sounds of running water. In Experiment 2, a test model of the specially designed sound masking speaker was used to present the masking noise. Further, sounds of tweeting birds and healing music were added to the mixed noise from Experiment 1 to create the three masking noise conditions. The average level of understanding for the mixed noise condition was 14%, while that for the bird and music conditions were 24% and 30% respectively. The higher understanding rates for the latter conditions were due to lower volume of the mixed white noise in order to keep the overall volume including the birds and music at 55dB. There were also significant effects of sentence type and reading voice gender, suggesting that auditory legibility does not solely depend on the speech-to-noise sound level ratio, but also on other variables, such as, predictability of the sentences, and clarity of the speech. Feedback at the end of the sessions revealed that the participants found mixed noise less irritating than pure white noise, and they preferred mixed noise with bird tweeting or music even better. Thus, it was concluded that mixed noise with occasional sounds of tweeting birds, was the most suitable masking sound for commercial use, being efficient and not unpleasant."
WAGLFOR
Artikel Jurnal  Universitas Indonesia Library
cover
Kulshreshtha, Manisha
"Dialect accent features for establishing speaker identity : a case study discusses the subject of forensic voice identification and speaker profiling. Specifically focusing on speaker profiling and using dialects of the Hindi language, widely used in India, the authors have contributed to the body of research on speaker identification by using accent feature as the discriminating factor. This case study contributes to the understanding of the speaker identification process in a situation where unknown speech samples are in different language/dialect than the recording of a suspect. The authors' data establishes that vowel quality, quantity, intonation and tone of a speaker as compared to Khariboli (standard Hindi) could be the potential features for identification of dialect accent."
New York: [Springer, ], 2012
e20418417
eBooks  Universitas Indonesia Library
cover
Lea, Wayne A.
:Englewood Cliffs, NJ : Prentice-Hall, 1980
621.380 412 LEA t
Buku Teks  Universitas Indonesia Library
<<   1 2 3 4 5 6 7 8 9 10   >>