Ryuki Tachibana

An audio watermarking method using a two-dimensional pseudo-random array

In this paper, we describe a multiple-bit audio watermarking method which is robust against wow-and-flutter, random sample cropping, and pitch shifting. Though these processings are easy to perform, they are difficult for audio watermarks to survive, because they introduce mis-synchronization between the embedded and detection watermarks. Our main ideas against these mis-synchronization attacks are a two-dimensional pseudo-random array (PRA), magnitude modification, and non-linear subbands. The embedding algorithm modifies the magnitudes of segmented areas in the time-frequency plane of the content, according to a two-dimensional pseudo-random array, while the detection algorithm correlates the magnitudes with the PRA. The two-dimensional array makes the watermark robust against cropping because, even when some portions of the content are heavily degraded, other portions of the content can match the PRA and contribute to watermark detection. Secondly, the magnitude modification enables detection even from displaced detection windows. This is because magnitudes are less influenced than phases by fluctuations of the analysis windows caused by random cropping. The last idea, wider bandwidths at higher frequencies, keeps the correspondence of the embedded and detection PRA even for pitch-shifted content. We theoretically and experimentally analyze the robustness of the proposed algorithm against a variety of signal degradations.

Audio watermarking method robust against time- and frequency-fluctuation

In this paper, we describe an audio watermarking algorithm that can embed a multiple-bit message which is robust against wow-and-flutter, cropping, noise-addition, pitch-shift, and audio compressions such as MP3. The algorithm calculates and manipulates the magnitudes of segmented areas in the time-frequency plane of the content using short-term DFTs. The detection algorithm correlates the magnitudes with a pseudo-random array that maps to a two-dimensional area in the time-frequency plane. The two-dimensional array makes the watermark robust because, even when some portions of the content are heavily degraded, other portions of the content can match the pseudo-random array and contribute to watermark detection. Another key idea is manipulation of magnitudes. Because magnitudes are less influenced than phases by fluctuations of the analysis windows caused by random cropping, the watermark resists degradation. When signal transformation causes pitch fluctuations in the content, the frequencies of the pseudo-random array embedded in the content shift, and that causes a decrease in the volume of the watermark signal that still correctly overlaps with the corresponding pseudo-random array. To keep the overlapping area wide enough for successful watermark detection, the widths of the frequency subbands used for the detection segments should increase logarithmically as frequency increases. We theoretically and experimentally analyze the robustness of proposed algorithm against a variety of signal degradations.

Watermarked Movie Soundtrack Finds the Position of the Camcorder in a Theater

In recent years, the problem of camcorder piracy in theaters has become more serious due to technical advances in camcorders. In this paper, as a new deterrent to camcorder piracy, we propose a system for estimating the recording position from which a camcorder recording is made. The system is based on spread-spectrum audio watermarking for the multichannel movie soundtrack. It utilizes a stochastic model of the detection strength, which is calculated in the watermark detection process. Our experimental results show that the system estimates recording positions in an actual theater with a mean estimation error of 0.44 m. The results of our MUSHRA subjective listening tests show the method does not significantly spoil the subjective acoustic quality of the soundtrack. These results indicate that the proposed system is applicable for practical uses.

Improving Audio Watermark Robustness Using Stretched Patterns against Geometric Distortion

One of the problems for audio watermarks is robustness against signal processing causing de-synchronization of the pseudo-random sequences. To tackle the problem, we previously introduced an audio watermarking method using a two-dimensional pseudo-random array, which is robust against pitch shifting and random stretching to some extent. In this paper, we explain a modification to the detection algorithm to improve the robustness against excessive distortion. The method uses multiple pseudo-random arrays each of which is stretched assuming a certain amount of distortion. Since most of the detection process for the multiple arrays is shared, the additional computational cost is limited.

Two-dimensional audio watermark for MPEG AAC audio

Since digital music is often stored in a compressed file, it is desirable that an audio watermarking method in a content management system handles compressed files. Using an audio watermarking method that directly manipulates compressed files makes it unnecessary to decompress the files before embedding or detection, so more files can be processed per unit time. However, it is difficult to detect a watermark in a compressed file that has been compressed after the file was watermarked. This paper proposes an MPEG Advanced Audio Coding (AAC) bitstream watermarking method using a two-dimensional pseudo-random array. Detection is done by correlating the absolute values of the recovered MDCT coefficients and the pseudo-random array. Since the embedding algorithm uses the same pseudo-random values for two adjacent overlapping frames and the detection algorithm selects the better frame in the two by comparing detected watermark strengths, it is possible to detect a watermark from a compressed file that was compressed after the watermark was embedded in the original uncompressed file. Though the watermark is not detected as clearly in this case, the watermark can still be detected even when the watermark was embedded in a compressed file and the file was then decompressed, trimmed, and compressed again.

Estimation of recording location using audio watermarking

In this paper, we propose a novel application of audio watermarking, estimation of recording location. The purpose of the paper is to determine the seat location in a theater at which a bootleg recording was made by using a digital video camera. In the proposed application, we embed different watermarks in the channels of the multi-channel sound of the movie. The multi-channel sound enters the air from multiple loudspeakers in a theater. If a monaural recording of the sound is made, the location of recording in the theater can be determined by detecting the multiple watermarks without any access to the host signals. The main idea is to use the time offsets of the watermark signals in the recorded signal. In simulation experiments, the recording locations can be determined within 1 m in almost all of the test cases. Although an experiment in a real environment resulted in more errors due to noise, these results successfully showed the potential applicability for practical use.

Determining Recording Location Based on Synchronization Positions of Audiowatermarking

In this paper, we propose a novel application of digital watermarking, determination of recording locations. This application enables us to determine the seat location in an auditorium where a recording was made. Precisely measured synchronization positions of the spread-spectrum watermarks are used for the determination. To avoid use of mismeasured synchronization positions, the algorithm discards synchronization positions with the corresponding normalized correlation values below a threshold. The experiments with our implementation resulted in accurate determinations; almost all of the locations can be determined within the error of 0.5 m. These experimental results successfully show the potential applicability of our application

Maximum-Likelihood Estimation of Recording Position Based on Audio Watermarking

In this paper, we propose a maximum-likelihood method for estimating a recording position inside several loudspeakers as a brand-new application of digital audio watermarking, which could be useful for protecting from illegal recordings of movies. An error model of times of arrival (TOAs) is utilized to estimate the recording position. The error model is constructed based on watermarking strengths, which are calculated by a watermark detection algorithm. This enables us to estimate the position if some of the TOAs are accurately measured. Experimental results indicate that the algorithm can estimate the recording position with the mean value of errors of 0.81 m.

Frame-level AnyBoost for LVCSR with the MMI Criterion

This paper propose a variant of AnyBoost for a large vocabulary continuous speech recognition (LVCSR) task. AnyBoost is an efficient algorithm to train an ensemble of weak learners by gradient descent for an objective function.We present a novel training procedure that trains acoustic models via the MMI criterion using data that is weighted proportional to the summation of the posterior functions of previous round of weak learners. Optimized for system combination by n-best ROVER at runtime, data weights for a new weak learner are computed as a weighted summation of posteriors of previous weak learners. We compare a frame-based version and a sentence-based version of our proposed algorithm with a frame-based AdaBoost algorithm. We will present results on a voice search task trained with different amounts of data with gains of 5.1% to 7.5% relative in WER can be obtained by three rounds of boosting.

Automatic music monitoring and boundary detection for broadcast using audio watermarking

An application of watermarking for automatic music monitoring of radio broadcasts is discussed. By embedding information into the music as a watermark before broadcasting it, it is possible to keep track of what music has been on the air at what time, and for how long. However, to effectively implement this application, the handling of content transitions is important, because the detection reliability deteriorates at the content boundaries. In this paper, a method of detecting content boundaries using overlapping detection windows is described. The most probable pattern of content transition is selected under the condition that detection results from multiple windows are available. The derived rules are represented using a finite state model, which is useful for detection in real time. Experimental results on FM radio broadcasts are also presented.