Search

Pitch is a critical cue for speech separation in humans’ auditory perception. Although the technology of tracking pitch in single-talker speech succeeds in many applications, it’s still a challenging problem to extract pitch information from speech mixtures in machine perception. In this paper, we aimed to combine speech separation and pitch tracking together to let them benefit from each other. A multi-task learning framework was proposed, in which a unified objective that considered both speech separation and pitch tracking was used, based on the utterance-level permutation invariant training (uPIT) as well as deep clustering (DPCL). In such framework, two tasks were optimized simultaneously and could benefit from each other through the sharing layers in the networks. Experimental results indicated the proposed multi-task framework outperformed the corresponding single-task framework, in terms of both speech separation and pitch tracking. The improvement was more significant for challenging same-gender mixtures.

Speaker segmentation is an essential part of any diarization system. Applications of diarization include tasks such as speaker indexing, improving automatic speech recognition (ASR) performance and making single speaker-based algorithms available for use in multi-speaker environments. This paper proposes a multiple hypothesis tracking (MHT) method that exploits the harmonic structure associated with the pitch in voiced speech in order to segment the onsets and end-points of speech from multiple, overlapping speakers. The proposed method is evaluated against a segmentation system from the literature that uses a spectral representation and is based on employing bidirectional long short term memory networks (BLSTM). The proposed method is shown to achieve comparable performance for segmenting overlapping speakers only using the pitch harmonic information in the MHT framework.

Speaker-independent multi-pitch tracking has been a long-standing problem in speech processing. In this study, we extend a recurrent neural network - factorial hidden Markov model (RNN-FHMM) framework, and use the utterance-level permutation invariant training (uPIT) criterion for multi-pitch tracking. Separated speech and label permutations from a speech separation uPIT-RNN have been further incorporated to improve pitch tracking performance. We evaluate our methods on the GRID database. Results indicate that the proposed speech separation - pitch tracking system with matched uPIT label permutations outperforms all other gender-dependent and speaker-independent multi-pitch trackers. The improvement is more significant for challenging same-gender mixtures.

To increase a density in hard disk drive, bit patterned media recording (BPMR) is regarded as a promising candidate for the next generation storage system. Reduction of spacing between bit islands is essential to increase the density in BPMR but causes inter-symbol interference (ISI) and inter-track interference (ITI). Islands can be arranged on a rectangular pattern (regular array) or packed in a staggered pattern (hexagonal lattice), depending upon the lithography approach adopted in BPMR. Since the staggered pattern can mitigate the effect of the ITI due to the presence of adjacent tracks, bit error rate (BER) performance of staggered pattern is better than that of rectangular pattern in previous study. In another work of channel modeling for staggered pattern, the staggered pattern with bit-aspect ratios (BAR $) =$ Track Pitch $(T_{z}) /$ Bit Length $(T_{x})$ of 2 performs better than the staggered pattern with BAR of 1. In this paper, we investigate the performance of BPMR island patterns according to the distances of track pitch and bit length, respectively. Fig. 1 presents BER performance of island arrangement according to areal densities of 2 Tb/in 2 and 3 Tb/in 2. When the areal density is 2 Tb/in 2, $(T_{x} \quad =15.5$ nm, $T_{z} \quad =21.0$ nm) provides best performance. Also, when the areal density is 3 Tb/in 2, $(T_{x} \quad =13.0$ nm, $T_{z} \quad =17.0$ nm) shows best performance. Fig. 2 illustrates BER performance in accordance with island arrangements. Case (1) represents the arrangement of islands that distributed over a rectangular array at $T_{x}$ of 18.0nm and $T_{z}$ of 18.0nm in 2 Tb/in 2 and $T_{x}$ of 14.5nm and $T_{z}$ of 14.5nm in 3 Tb/in 2. Case (2) represents the arrangement of islands that distributed over a staggered pattern at $T_{x}$ of 21.0nm and $T_{z}$ of 15.5nm in 2 Tb/in 2 and $T_{x}$ of 17.0nm and $T_{z}$ of 13.0nm in 3 Tb/in 2. Case (3) represents the arrangement of islands that distributed over a staggered pattern at $T_{x}$ of 18.0nm and $T_{z}$ of 18.0nm in 2 Tb/in 2 and $T_{x}$ of 14.5nm and $T_{z}$ of 14.5nm in 3 Tb/in 2. Case (4) represents the arrangement of islands that distributed over a staggered pattern at $T_{x}$ of 15.5nm and $T_{z}$ of 21.0nm in 2 Tb/in 2 and $T_{x}$ of 13.0nm and $T_{z}$ of 17.0nm in 3 Tb/in 2. Since the influence of adjacent tracks of Case (4) is less than that of Case (1), Case (2), and Case (3), i.e., the effect of ITI is reduced, Case (4) shows the best performance in both 2 Tb/in 2 and 3 Tb/in 2.

Pitch estimation has been of great interest for several decades due to many important audio applications, such as music transcription, source separation, and speech coding. There are several approaches in the literature for estimating pitch, many of which make use of short-time spectrum analysis. A recently proposed algorithm, namely the PEFAC algorithm, performs pre-enhancement of speech components in the short time spectrum to yield a robust pitch estimation. This pre-enhancement procedure is based on a function that outlines the spectral envelope of human speech in the universal sense. In this paper, we propose to overcome some limitations of the PEFAC algorithm by employing an alternative enhancement procedure, which uses an estimation of the individual spectral envelope instead of using a universal function. This approach allows better correspondence to the specific speaker's spectral features. Experimental results show that the proposed algorithm outperforms the original PEFAC algorithm, especially in hard conditions such as low SNR and transient noise.

A reliable estimate of pitch in noisy speech is crucial for many speech applications. In this paper, we propose to use speaker-dependent (SD) deep neural networks (DNNs) to model the harmonic patterns of each speaker. Specifically, SD-DNNs take spectral features as input and estimate probabilistic pitch states at each time frame. We investigate two methods for SD-DNN training. The first one is direct training when speaker-dependent data is sufficient. The second one is speaker adaptation of a speaker-independent (SI) DNN with limited data. The Viterbi algorithm is then used to track pitch through time. Experiments show that both training methods of SD-DNNs outperform an SI-DNN based system as well as a state-of-the-art pitch tracking algorithm in all SNR conditions.

Pitch is an important characteristic of speech and is useful for many applications. However, pitch determination in noisy conditions is difficult. In this paper, we propose a supervised learning algorithm to estimate pitch using a convolutional neural network (CNN). Specifically, we use a CNN for pitch candidate selection, and dynamic programming for pitch tracking. Our experimental results show that the proposed method can obtain accurate pitch estimation and they show good generalization ability to new speakers and noisy conditions. We credit the success to the use of CNN, which is suitable for modeling the shift-invariant spectral feature for pitch detection.

Two robust controllers, a linear disturbance estimator (DE) based controller and a standard integral controller, are respectively designed for multiple 3-DOF helicopters to achieve attitude-trajectory synchronization in both elevation and pitch tracking. These two controllers are shown to have similar structure but with different tuning parameters. It is further discovered that DE benefits from a single-parameter tuning for small tracking error while the other controller requires multiple parameters. Experimental investigations demonstrate the effectiveness of both controllers as well as the differences in tracking accuracy owing to different controller parameters.

Hard Disk Drive (HDD) Industry is approaching the end of the conventional perpendicular magnetic recording (PMR) technology due to the limitations of the super paramagnetic effect [1-3]. Shingled magnetic recording (SMR) is the current likely alternative to extend areal density beyond 1 Tb/ in2. SMR is feasible [4-7] and the HDD industry has delivered the drives to the market after extensive research around the globe. Earlier studies by many researchers thought wide writers would provide good writability and would no longer be an issue in SMR. However, in our study, we reveal that there exists an optimal write width for SMR for a given writer design. In this work, we aim to determine the optimal writer width and the shingling track pitch conditions.

This paper describes an approach that estimates the circularity coefficients of multiple underlying components within complex subbands of an additive mixture of voiced speech and noise via the strong uncorrelating transform (SUT). For the SUT to be effective, the latent source signals must have unique nonzero circularity coefficients; this requirement is satisfied by using narrow filters to impose a degree of noncircularity upon what would typically be circular noise. The circularity coefficient estimates are then used for voice activity detection, pitch tracking, and enhancement.

A deep neural network (DNN) based classifier achieved 27.38% frame error rate (FER) and 15.62% segment error rate (SER) in recognizing five tonal categories in Mandarin Chinese broadcast news, based on 40 mel-frequency cepstral coefficients (MFCCs). The same architecture scored substantially lower when trained and tested with F 0 and amplitude parameters alone: 40.05% FER and 22.66% SER. These results are substantially better than the best previously-reported results on broadcast-news tone classification [1] and are also better than a human listener achieved in categorizing test stimuli created by amplitude- and frequency-modulating complex tones to match the extracted F 0 and amplitude parameters.

This paper specifically targets violin learners who are working on their pitch accuracy. We employ a pitch tracking algorithm to extract the pitch played. Through volume thresholding and region detection, only parts of frames are processed. So our system can provide real-time feedback to show violin learners whether they played the right pitch. The system also provides the major scales and arpeggio scores as teaching materials, and violin learners can choose different tempos to practice, depending on their level. The user-friendly system interface allows violin learners to easily perceive the pitch differential between the pitch of the target note and the pitch played, allowing users to precisely adjust their playing. The statistical feedback records progress and analyzes error patterns, enabling violin teachers to evaluate student progress precisely, and correct common error patterns effectively.

In this paper, we propose a hybrid method for singing pitch extraction from polyphonic audio music. We have observed several kinds of pitch errors made by a previously proposed algorithm based on trend estimation. We also noticed that other pitch tracking methods tend to have other types of pitch error. Then it becomes intuitive to combine the results of several pitch trackers to achieve a better accuracy. In this paper, we adopt 3 methods as a committee to determine the pitch, including the trend-estimation-based method for forward and backward signals, and training-based HMM method. Experimental results demonstrate that the proposed approach outperforms the best algorithm for the task of audio melody extraction in MIREX 2010.

This chapter contains sections titled: Introduction Disk Drive Design Considerations Track Density and Storage Capacity of a Disk Track Misregistration Disk Drive Servo Sector, or Embedded, Servo Integration of Head, Medium, and Channel Track Profile and Microtrack Profile Off-track or OI Measurement Estimation of the System Error Rate Track Pitch Determination, "Squeeze," and the "747" Curve Application of Integration Procedures Window Margin Integration Procedures This chapter contains sections titled: References ]]>

This paper proposes a robust and accurate multi-pitch estimation method for multiple voices. This method is based on the spectral analysis of the mixture sound multi-scale product. The multi-scale product (PM) consists of making the product of wavelet transform coefficients. The wavelet used is the quadratic spline function. Simulation results showed that the proposed method can robustly estimate FOs for clean speech and for speech mixed with various interferences.

Estimating multiple pitch frequencies of concurrent speech sources from a single-microphone input is essential to speech separation. Nevertheless, pitch cues of individual sources are weakened by each other, making the estimation unreliable. This paper presents a pitch tracking method that incorporated in a model-based separation framework. Multiple pitch estimation is simplified into single pitch estimation by segregating the source envelope from mixture spectrum with statistics of familiar speech patterns. Comprehensive experiments have compared the proposed tracking method with a recently reported multiple pitch estimator and its modified version equipped with ideal pitch cues. Lower estimation errors are achieved. Furthermore, this approach is applicable to other model-based frameworks as well.

Fundamental frequency contours for speech, as obtained by common pitch tracking algorithms, contain a great deal of fine detail that is unlikely to hold much perceptual significance for listeners. In our experiments, a radically reduced pitch contour consisting of a single linear segment for each syllable was found to judged as equally natural as the original pitch track by listeners, based on high-quality analysis- synthesis. We describe the algorithms both for segmenting speech into syllables based on fitting Gaussians to the energy envelope, and for approximating the pitch contour by independent linear segments for each syllable. We report our web-based test in which 40 listeners compared the stylized pitch contour resyntheses to equivalent resyntheses based on the original pitch track, and also to pitch tracks stylized by the existing Momel algorithm. Listeners preferred the original pitch contour to the linear approximation in only 60% of cases, where 50% would indicate random guessing. By contrast, the original was preferred over Momel in 74% of cases.

In this paper, we describe the BBN 2007 Mandarin speech-to-text system developed for the GALE Evaluation 2007. In comparison to the BBN 2006 Mandarin system, we achieved 25% relative reduction in character error rate on the most important test sets. The utilization of all available training data provided the largest contribution to the improvement. The use of a better pitch tracking algorithm also contributed significantly, while system combination made some noticeable improvement too.

Fabrication of ultrahigh packed pit and dot arrays have been studied using conventional electron beam (EB) writing, and positive and negative EB resists, ZEP520 and calixarene, respectively. Using fine electron beam with high probe current and very thinner resists, we demonstrate that the negative resist has a potential to achieve an ultrahigh density storage with both bit pitch (BP) and track pitch (TP) of 1 trillion bits per inch2(Tb/in2) storage technology.

The paper proposed a novel super-RENS ROM disc with a narrow track pitch using a group tracking technique, GTS-super-RENS ROM disc. The group tracking technique enabled to increase capacity in the radial direction up to 1.5 times. Using GTS method, a -20 dB cross-talk from the neighbor pit trains was obtained. It was confirmed that capacity of GTS-super-RENS ROM disc will become a promising method to further increase capacity in super-RENS disc world

This paper presents a simple practical look-back and look-forth pitch tracking and smoothing algorithm based on dynamic programming approach. The experimental results show that this tracking and smoothing algorithm can improve the accuracy of pitch estimation when it is incorporated into the multiband excitation time domain autocorrelation pitch estimation.

This paper reports on the structure and performance of multichannel write and giant magnetoresistive (GMR) heads for a helical-scan tape system. The four-channel write head was fabricated with four layers stacked and shifted by a track pitch. The eight-channel read head was also stacked four layers, each of which has two spin-valve elements. The write head and read head were assembled into a commercial drive. Four channels are written in one revolution without a guard band and each of the eight channels is read, in half of the track pitch. The feasibility of a recording density over 4.3 Gb/in2 was established, which would make possible a tape system of over 1 TB using an 8 mm cassette

A new type of moving-slider microactuator for high track pitch HDD was investigated. The new microactuator uses PZT ceramic as a source of actuation. Since PZT can generate high force, the microactuator can use a very stiff spring while achieving the required stroke. A microactuator has a stroke amplification mechanism, so that the small stroke generated by PZT ceramic is amplified. In order to take advantage of superior material property of bulk PZT ceramic, a "hybrid" approach was employed: bulk PZT ceramic and microactuator structure is separately processed/fabricated, then, these two are assembled at the final step of fabrication process. A microactuator and PZT ceramic was successfully fabricated, assembled, and tested. Frequency response measurement showed excellent performance of this microactuator.

Magnetically induced superresolution (MSR) technology expands the readout resolution limit of the optical disk system. A MSR-by-RAD (rear aperture detection) disk, was used to demonstrate a crosstalk level of less than -44 dB at a track pitch of 0.8 mu m and a carrier-to-noise ratio of 45 dB at a mark length of 0.4 mu m. The results show the feasibility of a magneto-optical disk system with a recording capacity six times larger than that of the current ISO disk system. >

Various kinds of DVD such as DVD-R/RW, +R/RW and -RAM (2.6 GB and 4.7 GB) are on the market and some next generation optical disks such as Blu-Ray disk are set to appear in the near future. The development of a read/write pick-up for the various kinds of disk is thus expected. In this paper, we propose a new tracking servo system named phase-shift differential push-pull (PS-DPP) method which is not dependent on track pitch and the position of the sub-beams. This system is applicable for all kinds of DVD. This paper describes the concept of PS-DPP, the design of the new inclined multidivision phase-shift grating, and the simulated and the experimental results of its push-pull offset compensation effect.

A next generation optical disk is required to have over 23 Gbyte recording capacity for over 2 hour storage of digital high-definition television (HDTV) data. Since track pitch becomes narrower for such a large capacity disk, a high resolution and high accuracy mastering system is required. To realize this objective, we have developed an electron beam recorder (EBR) and have been studying fabrication of master stampers for next generation optical disks and for future optical disks (Y. Kojima et al, Jpn. J. Appl. Phys. vol. 37, p. 2137, 1998; Y. Wada et al, ibid., vol. 40, p. 1653, 2001). In this paper, we report an investigation of recording accuracy in the EBR and a trial for improvement.

The read-write characteristics of a land and groove disk have been examined using a blue-laser diode, high NA objective lens and the PRML method. The wide tilt and defocus margins have been obtained and no deterioration by CW has been observed at a data bit length of 0.0987 /spl mu/m and a track pitch of 0.29 /spl mu/m. These experimental results certify the feasibility of a rewritable optical disk with a capacity of 30 GB/side.

The authors present an improved version of a previous 8-kb/s low-delay VXC (vector excitation coding) algorithm, which has been enhanced to maintain high-quality results under clean conditions while adding robustness to channel errors. Periodic resetting of the pitch predictor makes the coder more robust to channel errors but sacrifices some of the advantages of interframe coding in clean channels. A novel adaptive mean-removed pitch tracking scheme was introduced that provides robustness to bit errors while avoiding significant quality degradation under error-free conditions. >

Engineering researchers at Loughborough University and hearing researchers at Essex University are working together to produce a real-time digital electronic pitch extraction system based on neuromorphic principles and implemented in digital technology. This paper reports on the VHDL simulations of the system which demonstrates good and robust pitch tracking across the human voice range.

Data storage technologies such as HDD and optical recordings are key to the information-oriented society, and the demand for their technological developments will be undoubtedly more strongly increasing. For mass-data storage applications, the capacity or areal density of drives will be more demanded. With increasing areal density, both track pitch and bit-length must be decreased. Yet, serious issues remain to be solved before any drives will be commercially available. These issues are discussed.

The pupil light distribution for a DVD-RAM type land groove optical disc storage media with a 0.5 /spl mu/m track spacing is discussed, along with the implementation of quadrant pupil detection (QPD) and tri-segmented pupil detection (TPD) crosstalk reduction schemes.

In this paper we compare four different pitch tracking algorithms - autocorrelation, cepstrum, harmonic product spectrum, and a new method based on the modulation spectrum - and investigate their fitness for noisy environments. From each tracker, possible ƒ 0 candidates in every time window are stored over a 10-frame interval, and the best contour is computed through a Viterbi search. The audio data comprises speech samples recorded in a moving car with a microphone setup commonly used for car speech applications like speakerphones. To see how performances deteriorate with increasing noise level, a second audio test set is prepared by blending clean speech recordings and noise at different signal-ta-noise ratios.

We present a robust algorithm for multi-pitch tracking of noisy speech. Our approach integrates an improved channel and peak selection method, a new integration method for extracting periodicity information across different frequency channels, and a hidden Markov model (HMM) for forming continuous pitch tracks, and as a result, our algorithm can reliably track single and double pitch tracks in a noisy environment. The proposed algorithm is evaluated on a database of speech utterances mixed with various interferences and the results show that our algorithm outperforms existing algorithms significantly.