Citation Context

...ailed analysis can be done to provide type specific information of the input signal, according to the requirements of the application. For voice and music input, techniques such as speech recognition =-=[32]-=-, speaker identification [33], or music genre classification [41] can be applied. The features we adopt are widely used by other audio and music classification [26]. More powerful yet computationally ...

Citation Context

...d with this in mind. The microphone on a phone is typically designed for capturing the human voice, not ambient sounds, and typically sample at 8 KHz. According to the NyquistShannon sampling theorem =-=[38]-=-, the microphone cannot capture information above 4 KHz, and, as a result, important information is lost, for example, high frequency components of music. Beyond sampling issues, mobile phones do not ...

Citation Context

...to perform the following three steps: (i) apply a hanning window to the frame, which suppresses the frame boundaries and thus reduces the known effect of Fast Fourier Transform (FFT) spectral leakage =-=[18]-=-; (ii) calculate the FFT spectrum of the frame; and finally, (iii) normalize the spectrum, treat it as a probability density function, and finally obtain the spectral entropy, Hf, by, Hf = − nX i=1 pi...

Citation Context

...] [22] and urban sensing [11] [10] [1] applications. Although early work [36] mentions the use of microphone as a clue to context, most work found in the literature focus on the camera, accelerometer =-=[6]-=-, and GPS [23] as sources of sensor data. There has been significant work on audio analysis and signal processing. The basic problem of sound classification has been as an active area of research [35]...

Citation Context

...r has complete control over how the results of classification are presented either in terms of being visualized on the phone screen or pushing them to external applications, such as, social networks. =-=[27]-=-. Note, that in our current implementation results can only be displayed locally on the user’s phone. Another pressing challenge when designing continuous sensing applications such as SoundSense is th...

Citation Context

...23] as sources of sensor data. There has been significant work on audio analysis and signal processing. The basic problem of sound classification has been as an active area of research [35] [34] [21] =-=[16]-=- including some of the challenges we overcome with Sound-concerns of communicating raw audio samples and features to central servers for retraining. In [39], the authors demonstrate a system able to ...

Citation Context

...ained mobile phones yet be effective and make an attempt to safeguard the privacy of users.SoundSense leverages the existing body of work on acoustic signal processing and classification (e.g., [24] =-=[34]-=- [21] [14]) and takes a systems design approach to realizing the first such system for a mobile phone. A key design goal of SoundSense is the scalability of classification to a large population. Speci...

Citation Context

...meters). These mobile devices are leading to the emergence of new applications in health care, gaming, social networks, and recreational sports - enabling people-centric sensing applications [11] [1] =-=[10]-=-. Perhaps one of the most overlooked “sensors”, available not only on top end phones but also on every mobile phone, is the microphone. Along with the camera, the microphone is one of the two most ubi...

Citation Context

...ification. The frame width (i.e., duration) needs to be short enough so that the audio content is stable and meanwhile long enough to capture the characteristics signature of the sound. Existing work =-=[26]-=- use frames that overlap each other since overlapping frames are able to capture subtle changes in the sounds more precisely. However, this can cause overlapping pieces of audio data to be processed m...

Citation Context

...ban sensing [11] [10] [1] applications. Although early work [36] mentions the use of microphone as a clue to context, most work found in the literature focus on the camera, accelerometer [6], and GPS =-=[23]-=- as sources of sensor data. There has been significant work on audio analysis and signal processing. The basic problem of sound classification has been as an active area of research [35] [34] [21] [16...

Citation Context

... provide type specific information of the input signal, according to the requirements of the application. For voice and music input, techniques such as speech recognition [32], speaker identification =-=[33]-=-, or music genre classification [41] can be applied. The features we adopt are widely used by other audio and music classification [26]. More powerful yet computationally demanding features, for examp...

Citation Context

...locations where the music are detected along with icons on the map that indicate the location of the sound. 7. RELATED WORK There is growing interest in mobile phones research for pervasive computing =-=[29]-=- [22] and urban sensing [11] [10] [1] applications. Although early work [36] mentions the use of microphone as a clue to context, most work found in the literature focus on the camera, accelerometer [...

Citation Context

... accelerometers). These mobile devices are leading to the emergence of new applications in health care, gaming, social networks, and recreational sports - enabling people-centric sensing applications =-=[11]-=- [1] [10]. Perhaps one of the most overlooked “sensors”, available not only on top end phones but also on every mobile phone, is the microphone. Along with the camera, the microphone is one of the two...

Citation Context

...ited range of frequencies, having a small value. Music often consists of a broader mixture of frequencies than voice and ambient sound. Normalized Weighted Phase Deviation. This feature introduced in =-=[15]-=- shows the phase deviations of the frequency bins in the spectrum weighted by their magnitude, where φ ′′ i nwpdf = nX i=1 p(i) · φ ′′ i (7) is the second derivative of the phase of ith frequency bin....

Citation Context

...lerometers). These mobile devices are leading to the emergence of new applications in health care, gaming, social networks, and recreational sports - enabling people-centric sensing applications [11] =-=[1]-=- [10]. Perhaps one of the most overlooked “sensors”, available not only on top end phones but also on every mobile phone, is the microphone. Along with the camera, the microphone is one of the two mos...

Citation Context

... phone, their environments and social events. This single modality is capable of supporting a diverse set of inferences, such as: conversation detection, activity recognition, location classification =-=[30]-=- [14], discovery of social network structure [13], and even classification of dietary intake of a person [2]. Moreover, the microphone is robust to a range of phone contexts (e.g., body position of th...

Citation Context

...s single modality is capable of supporting a diverse set of inferences, such as: conversation detection, activity recognition, location classification [30] [14], discovery of social network structure =-=[13]-=-, and even classification of dietary intake of a person [2]. Moreover, the microphone is robust to a range of phone contexts (e.g., body position of the device). Many modalities are rendered useless b...

Citation Context

...cate the location of the sound. 7. RELATED WORK There is growing interest in mobile phones research for pervasive computing [29] [22] and urban sensing [11] [10] [1] applications. Although early work =-=[36]-=- mentions the use of microphone as a clue to context, most work found in the literature focus on the camera, accelerometer [6], and GPS [23] as sources of sensor data. There has been significant work ...

Citation Context

... mobile phones yet be effective and make an attempt to safeguard the privacy of users.SoundSense leverages the existing body of work on acoustic signal processing and classification (e.g., [24] [34] =-=[21]-=- [14]) and takes a systems design approach to realizing the first such system for a mobile phone. A key design goal of SoundSense is the scalability of classification to a large population. Specifical...

Citation Context

...e, their environments and social events. This single modality is capable of supporting a diverse set of inferences, such as: conversation detection, activity recognition, location classification [30] =-=[14]-=-, discovery of social network structure [13], and even classification of dietary intake of a person [2]. Moreover, the microphone is robust to a range of phone contexts (e.g., body position of the dev...

Citation Context

...e classifier evolved out of a study investigating the use of several other classification algorithms that have proven to be effective in the audio processing including: Gaussian Mixture Models (GMMs) =-=[40]-=- [26], knearest neighbors(KNN) [30], decision trees [44] and Hidden Markov Models (HMMs) [44]. We construct classifiers using all of these algorithms with an identical set of features (see Section 4.2...

Citation Context

... intra-category classifier – the gender classifier. Future work will extend this initial implementation. The classifier is fairly simple in comparison to other examples found in the literature (e.g., =-=[37]-=- reports 90% accuracy) so we do not expect high accuracy for this classifier. We perform an experiment using the voice clips in the test data set. According to this data set, we achieve 72% classifica...

Citation Context

...inferences, such as: conversation detection, activity recognition, location classification [30] [14], discovery of social network structure [13], and even classification of dietary intake of a person =-=[2]-=-. Moreover, the microphone is robust to a range of phone contexts (e.g., body position of the device). Many modalities are rendered useless by particular phone context (e.g., the camera when the phone...

Citation Context

...the second derivative of the phase of ith frequency bin. Usually the ambient sound and music will have a smaller phase devation than voice. Relative Spectral Entropy (RSE). This feature introduced in =-=[7]-=- differentiates speech and other sounds. However, in [7] RSE is defined as the KL (Kullback-Leibler) divergence between the frequency domain of the current window of frames and previous 500 frames. Gi...

Citation Context

...ambient sound learning algorithm impacts the computational costs and classification accuracy. We examine this tradeoff with a number of experiments on the Intel MSP (Mobile Sensing Platform) data set =-=[12]-=- collected as part of earlier work. The MSP data set contains fifty-two 25minute-long sound clips collected by the MSP device, which senses audio signals with a microphone similar to those on the mobi...

Citation Context

...nsupervised Adaptive Classifier performance make the labels understandable to themsleves and possibly others people. Sourcing labels from user populations is a complex research topic in its own right =-=[19]-=-. While this is an interesting systems issue it is out of scope of this paper, but certainly a topic we intend to study in future. Using the MSP data set [12], we evaluate the effectiveness of the uns...

Citation Context

...rform queries against this log of sound events determining for instance how much time they spent in their cars, etc. There has been a number of applications similar to this proposed in the literature =-=[42]-=- [20]. To experiment with this application we had a single participant carry an iPhone to collect raw audio data over theEvent Voice Music Driving Misc. Silence 11:00 12:00 1:00 2:00 3:00 4:00 5:00 6...

Citation Context

...ture extraction and scalable classification that is capable of running online on resource limited phones. This has not be achieved before in the literature and distinguishes our contribution. In [24] =-=[25]-=- the authors use a HMM based strategy capable of classifying 10 environments with good performance using samples of only 3 second duration. This framework provides a way to build new classifiers. Howe...

Citation Context

... queries against this log of sound events determining for instance how much time they spent in their cars, etc. There has been a number of applications similar to this proposed in the literature [42] =-=[20]-=-. To experiment with this application we had a single participant carry an iPhone to collect raw audio data over theEvent Voice Music Driving Misc. Silence 11:00 12:00 1:00 2:00 3:00 4:00 5:00 6:00 7...

Citation Context

...onstrained mobile phones yet be effective and make an attempt to safeguard the privacy of users.SoundSense leverages the existing body of work on acoustic signal processing and classification (e.g., =-=[24]-=- [34] [21] [14]) and takes a systems design approach to realizing the first such system for a mobile phone. A key design goal of SoundSense is the scalability of classification to a large population. ...

Citation Context

...els of details to be inferred from the sound sample, whose category is already classified. For instance, beyond just inferring the sound contains music, further analysis can assess the genre of music =-=[41]-=-. Performing these types of finer grain distinctions can only happen after the type of the sound is known. In this stage, we can apply techniques from prior work, and in particular make use of work th...

Citation Context

... an active area of research [35] [34] [21] [16] including some of the challenges we overcome with Sound-concerns of communicating raw audio samples and features to central servers for retraining. In =-=[39]-=-, the authors demonstrate a system able to recognize a number of sounds using mobile devices but again this system makes extensive use of off device computational resources offered by back-end infrast...

Citation Context

...n gender classifier ambient sound analyser serialization m / f sound type user label user interface sqlite data base Figure 5: Implementation component diagram. is a mixture of C, C++ and objective C =-=[3]-=-. Our signal processing and core classification algorithms are mainly written in C and C++. Objective C is necessary to build an Apple iPhone application which allows us to access hardware (e.g., micr...