Find research datasets worth reusing

TIMIT-TTS: a Text-to-Speech Dataset for Synthetic Speech Detection

With the rapid development of deep learning techniques, the generation and counterfeiting of multimedia material are becoming increasingly straightforward to perform. At the same time, sharing fake content on the web has become so simple that malicious users can create unpleasant situations with minimal effort. Also, forged media are getting more and more complex, with manipulated videos (e.g., deepfakes where both the visual and audio contents can be counterfeited) that are taking the scene over still images. The multimedia forensic community has addressed the possible threats that this situation could imply by developing detectors that verify the authenticity of multimedia objects. However, the vast majority of these tools only analyze one modality at a time. This was not a problem as long as still images were considered the most widely edited media, but now, since manipulated videos are becoming customary, performing monomodal analyses could be reductive. Nonetheless, there is a lack in the literature regarding multimodal detectors (systems that consider both audio and video components). This is due to the difficulty of developing them but also to the scarsity of datasets containing forged multimodal data to train and test the designed algorithms. In this paper we focus on the generation of an audio-visual deepfake dataset. First, we present a general pipeline for synthesizing speech deepfake content from a given real or fake video, facilitating the creation of counterfeit multimodal material. The proposed method uses Text-to-Speech (TTS) and Dynamic Time Warping (DTW) techniques to achieve realistic speech tracks. Then, we use the pipeline to generate and release TIMIT-TTS, a synthetic speech dataset containing the most cutting-edge methods in the TTS field. This can be used as a standalone audio dataset, or combined with DeepfakeTIMIT and VidTIMIT video datasets to perform multimodal research. Finally, we present numerous experiments to benchmark the proposed dataset in both monomodal (i.e., audio) and multimodal (i.e., audio and video) conditions. This highlights the need for multimodal forensic detectors and more multimodal deepfake data. <ul> <li>For the initial version of TIMIT-TTS v1.0 <ul> <li>Arxiv: https://arxiv.org/abs/2209.08000</li> <li>TIMIT-TTS Database v1.0: https://zenodo.org/record/6560159</li> </ul> </li> </ul>

opencc-by-4.0Aug 2022View details →

Hi, KIA: A Speech Emotion Recognition Dataset for Wake-Up Words

Hi,KIA dataset is a shared short Wakeup Word database focusing on perceived emotion in speech The dataset contains 488 Wakeup Word speech.  For more detailed information about the dataset, please refer to our paper: Hi, KIA: A Speech Emotion Recognition Dataset for Wake-Up Words File Description <ul> <li>wav/: wav files. <ul> <li>Filename f`{gender}_{pid}_{scene}_{trial}_{emotion}.wav` The first letter was used to express emotion.  </li> </ul> </li> <li>annotation/: Information related to annotation and human validation of the entire speech</li> <li> split: 8fold data split with {train, valid, test}.csv  </li> <li> handcraft: Features used for data EDA and baseline performance </li> <li> best_weights: wav2vec2.0 context network finetuning weights for re-implementation. Due to file size, we attach only fold M1, F5 </li> </ul>   Reference <ul> </ul> Hi, KIA: A Speech Emotion Recognition Dataset for Wake-Up Words [[ArXiv](https://arxiv.org/abs/2211.03371)] ``` @inproceedings{kim2022hi,   title={Hi, KIA: A Speech Emotion Recognition Dataset for Wake-Up Words},   author={Taesu Kim, SeungHeon Doh, Gyunpyo Lee, Hyung seok Jun, Juhan Nam, Hyeon-Jeong Suk},   booktitle={Proceedings of the 14th Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA)},   year={2022} } ```

Non-acoustic speech sensing system based on flexible piezoelectric

The non-acoustic speech sensing system based on flexible piezoelectric is designed to satisfy specific needs around testing device models (in high-noise, complex environments). The system collected vibration signals from the jaws of six males and five females containing ten different control commands at 90 dB of background noise. The dataset is reliable with high intelligibility and is able to achieve 93.7% recognition accuracy by calculation. In general, this paper provides a non-acoustic speech dataset for Mandarin, including the parts collected, the number of people collected, and the environment. The dataset is available at: https://doi.org/10.5281/zenodo.7095762 The data descriptor paper with details of data collection and cleaning process is under submission. For proper citation of the manuscript, please refer to the latest version of this dataset which includes the details. This dataset and its descriptor paper were created by: Shiji Yuan, Ying Sun, Dezhi Zheng, Xinlei Chen,Ying Ding, Shuai Wang, Shangchun Fan For questions or suggestions, please e-mail Dezhi Zheng <zhengdezhi@buaa.edu.cn> Description: Ten common words were chosen as the core of the vocabulary in this dataset. These ten command words can be used for commands in IoT or robotics applications: "forward", "backward", "right", "left", "stop", "up", "down", "draw", "drop", and "reset". The recording was carried on by software named Adobe Audition 2022. We set monophonic recording, 16-bit storage format, and 16 kHz sampling frequency before recording and saved the recorded voice in wav format.  The dataset is provided with two storage rules, which are stored by subject number and command number as classification. In the first rule, the speech data of 11 subjects were stored in different folders with the subject serial number as the folder name. Each folder contains subfolders categorized by command. In the second rule, the speech data of ten commands are stored in different folders, and the names of the folders are the command contents.  The subject number, command number and record order are given for each data entry. For example, the data obtained when subject 1 recorded command 10 for the first time was labeled as "1-10_1". After the data collection process, a filtering algorithm for automatic detection of low non-acoustic speech data was designed to remove problematic data that were very short or very quiet.The script of the data filtering algorithm is provided in this repository.   For specific detail of the data filtering process, please refer to the script (speech data filtering algorithm in MATLAB) in this repository and the data descriptor paper. The dataset in this repository is the processed version. The raw dataset and removed audio files are not included in this repository. File list: Non-acoustic Speech Dataset.zip speech data filtering algorithm.zip Readme.txt

Non-acoustic Speech Dataset

Non-acoustic speech sensing system based on flexible piezoelectric Version 1.0.0                                       This Read_Me.txt file briefly describes the non-acoustic speech dataset and instructions to access it.  The non-acoustic speech sensing system based on flexible piezoelectric is designed to satisfy specific needs around testing device models (in high-noise, complex environments). The system collected vibration signals from the jaws of six males and five females containing ten different control commands at 90 dB of background noise. The dataset is reliable with high intelligibility and is able to achieve 93.7% recognition accuracy by calculation. In general, this paper provides a non-acoustic speech dataset for Mandarin, including the parts collected, the number of people collected, and the environment. The dataset is available at: https://10.5281/zenodo.7090120 The data descriptor paper with details of data collection and cleaning process is under submission. For proper citation of the manuscript, please refer to the latest version of this dataset which includes the details. This dataset and its descriptor paper were created by: Shiji Yuan, Ying Sun, Dezhi Zheng, Xinlei Chen, Ying Ding,Shuai Wang, Shangchun Fan For questions or suggestions, please e-mail Dezhi Zheng <zhengdezhi@buaa.edu.cn> Description: Ten common words were chosen as the core of the vocabulary in this dataset. These ten command words can be used for commands in IoT or robotics applications: "forward", "backward", "right", "left", "stop", "up", "down", "draw", "drop", and "reset". The recording software is Adobe Audition2022,which adopts monophonic recording, 16-bit storage format, 16 kHz sampling frequency, and the recorded voice is saved in wav format. The dataset is provided with two storage rules, which are stored by subject number and corpus number as classification. In the first rule, the speech data of 11 subjects were stored in different folders with the subject serial number as the folder name. Each folder contains subfolders categorized by corpus. In the second rule, the speech data of ten corpus are stored in different folders, and the names of the folders are the corpus contents. The subject number, corpus number and record order are given for each data entry. For example, the data obtained when subject one recorded corpus 10 for the first time was labeled as 1-10_1. After the data collection process, a filtering algorithm for automatic detection of low non-acoustic speech data is designed to remove problematic data that are very short or very quiet. The script of the data filtering algorithm is provided in this repository.   For specific detail of the data filtering process, please refer to the script (speech data filtering algorithm in MATLAB) in this repository and the data descriptor paper. The dataset in this repository is the processed version. The raw dataset and removed audio files are not included in this repository. File list: Non-acoustic Speech Dataset.zip speech data filtering algorithm.zip Readme.txt

Non-acoustic speech sensing system based on flexible piezoelectric

The non-acoustic speech sensing system based on flexible piezoelectric is designed to satisfy specific needs around testing device models (in high-noise, complex environments). The system collected vibration signals from the jaws of six males and five females containing ten different control commands at 90 dB of background noise. The dataset is reliable with high intelligibility and is able to achieve 93.7% recognition accuracy by calculation. In general, this paper provides a non-acoustic speech dataset for Mandarin, including the parts collected, the number of people collected, and the environment. The dataset is available at: https://doi.org/10.5281/zenodo.7185663 The data descriptor paper with details of data collection and cleaning process is under submission. For proper citation of the manuscript, please refer to the latest version of this dataset which includes the details. This dataset and its descriptor paper were created by: Shiji Yuan, Ying Sun, Shuai Wang, Xinlei Chen,Ying Ding,Dezhi Zheng , Shangchun Fan For questions or suggestions, please e-mail Shuai Wang <wangshuai@buaa.edu.cn> Description: Ten common words were chosen as the core of the vocabulary in this dataset. These ten command words can be used for commands in IoT or robotics applications: "forward", "backward", "right", "left", "stop", "up", "down", "draw", "drop", and "reset". The recording was carried on by software named Adobe Audition2022. We set monophonic recording, 16-bit storage format, and 16 kHz sampling frequency before recording and saved the recorded voice in wav format.  The dataset is provided with two storage rules, which are stored by subject number and command number as classification. In the first rule, the speech data of 11 subjects were stored in different folders with the subject serial number as the folder name. Each folder contains subfolders categorized by command. In the second rule, the speech data of ten commands are stored in different folders, and the names of the folders are the command contents. The subject number, command number and record order are given for each data entry. For example, the data obtained when subject 1 recorded command 10 for the first time was labeled as "1-10_1". After the data collection process, a filtering algorithm for automatic detection of low non-acoustic speech data was designed to remove problematic data that were very short or very quiet.The script of the data filtering algorithm is provided in this repository.   For specific detail of the data filtering process, please refer to the script (speech data filtering algorithm in MATLAB) in this repository and the data descriptor paper. The dataset in this repository is the processed version. The raw dataset and removed audio files are not included in this repository. File list: Non-acoustic Speech Dataset.zip speech data filtering algorithm.zip Readme.txt

opencc-by-4.0Apr 2019View details →

Raw Data from - Speech-ABRs in Cochlear Implant Recipients: Feasibility Study

Speech-ABRs in Cochlear Implant Recipients: Feasibility Study Ghada BinKhamis, Emanuele Perugia, Martin O’Driscoll, Karolina Kluk Please site the paper when using this dataset (DOI: 10.1080/14992027.2019.1619100)   Participant Recordings (BinKhamis et al Speech_ABR in CI raw data.zip) <ul> <li>Description of raw EEG (speech-ABR) data main folder, subfolders, and raw EEG files <ul> <li>Folder Information <ul> <li>Main folder: Contains 12 subfolders with the raw data from 12 participants</li> <li>Subfolders: Each subfolder contains raw EEG recordings from one participant, subfolders names are:</li> <li>CI1, CI2, CI3, CI4, CI5, CI6, CI7, CI8, CI9, CI10, CI11, CI12</li> </ul> </li> <li>Each participant subfolder contains four ‘.mat’ files: <ul> <li>Each ‘.mat’ file starts with the participant ID <ul> <li>CI1, CI2, CI3, CI4, CI5, CI6, CI7, CI8, CI9, CI10, CI11, CI12</li> </ul> </li> <li>Followed by the stimulus: 40 da</li> <li>Followed by the stimulus polarity <ul> <li>Pos for positive/standard</li> <li>Neg for negative (reversed polarity stimulus)</li> </ul> </li> <li>Followed by the ipsilateral/implanted ear <ul> <li>R for right ear</li> <li>L for left ear</li> </ul> </li> <li>Finally the recording number for that polarity <ul> <li>1 is the first recording</li> <li>2 is the second recording</li> </ul> </li> </ul> </li> <li>Example ‘.mat’ file name: <ul> <li>CI1 40 da Neg R1.mat: Participant number 1, reversed polarity stimulus, implanted ear – right ear, recording number one</li> <li>CI9 40 da Pos L2.mat: Participant number 9, standard/positive polarity stimulus, implanted ear – left ear, recording number two</li> </ul> </li> <li>Description of ‘.mat’ files that can be accessed and processed using MATLAB (MathWorks): Each ‘.mat’ file is a structure that contains the following fields: <ul> <li>The first nine fields are informational, for example: <ul> <li>xunits: ‘s’ indicates that the recording time window is in seconds, conversion to milliseconds would be required to plot the data in milliseconds</li> <li>start: ‘0’ indicates that both stimulus and recording start at 0 seconds</li> <li>points: 13750 is the number of sample points</li> <li>chans: 2 is the number of channels (channel 2 is the ipsilateral channel for participants with a Right CI, and channel 1 is the ipsilateral channel for participants with a Left CI)</li> <li>frames: 2500 is the number of epochs (repetitions)</li> </ul> </li> <li>The last filed ‘values’ is what contains the raw EEG data (13750x2x2500) <ul> <li>13750 is the number of samples</li> <li>2 is the number of channels (channel one is recorded from the left ear lobe (A1) and channel two is from the right ear lobe (A2))</li> <li>2500 is the number of epochs</li> </ul> </li> </ul> </li> </ul> </li> </ul> Date of data collection: September 2017   Artificial-head artefact Recordings (BinKhamis et al Speech_ABR in CI artefact data.zip) <ul> <li>This folder contains six artificial-head artefact recordings <ul> <li>CI07_Artefact_40da_neg.mat <ul> <li>Artefact based on closest approximation to CI07's MAP</li> <li>Recorded in response to a reversed polarity (neg) 40 ms [da]</li> </ul> </li> <li>CI07_Artefact_40da_pos.mat <ul> <li>Artefact based on closest approximation to CI07's MAP</li> <li>Recorded in response to a standard polarity (pos) 40 ms [da]</li> </ul> </li> <li>CI08_Artefact_40da_neg.mat <ul> <li>Artefact based on closest approximation to CI08's MAP</li> <li>Recorded in response to a reversed polarity (neg) 40 ms [da]</li> </ul> </li> <li>CI08_Artefact_40da_pos.mat <ul> <li>Artefact based on closest approximation to CI08's MAP</li> <li>Recorded in response to a standard polarity (pos) 40 ms [da]</li> </ul> </li> <li>CI09_Artefact_40da_neg.mat <ul> <li>Artefact based on closest approximation to CI09's MAP</li> <li>Recorded in response to a reversed polarity (neg) 40 ms [da]</li> </ul> </li> <li>CI09_Artefact_40da_pos.mat <ul> <li>Artefact based on closest approximation to CI09's MAP</li> <li>Recorded in response to a standard polarity (pos) 40 ms [da]  </li> </ul> </li> </ul> </li> <li>Artefact recordings are in '.mat' format (MATLAB, MathWorks) <ul> <li>Each '.mat' file is a 10988x2500 matrix <ul> <li>10988 is the number of sample points</li> <li>2500 is the number of epochs (repetitions)</li> </ul> </li> </ul> </li> </ul> Artefacts recorded in: June 2018

Figure 10. Conceptual diagram illustrating vector quantization codeword formation-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

After the conversion process converts the analogue speech signal into a series words, then converting recognized word to facial animation based on VRML.

Figure 9. Mel Cepstral Coefficients in time domain-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

In this section, Feature matching Algorithm has been discussed. The goal of feature matching is to classify objects into one of a number of categories or classes. In this project, Vector Quantization approach will be used and the best matching result will be the desired voice.

Figure 6. Change the frequency at Hertz scale to Mel Scale.-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Following computing the spectrum power and applying the above equation on frequency axis, some mediating filters equal to identical overlapping are applied on the scaled spectrum and each filters energy is computed as particularity. This is because conception of a particular frequency by the auditory system is affected by a critical band of frequencies surrounding it. Number of filters is usually between 20 and 30. Logarithmic non linear change operations on obtained particulates for adjusting amplified of particularities and their important though coordinating them with the structure of the auditory system following computing energy of each filter is done as follows: in the following equation F1 is filter in I th, and e (i) is logarithm of energy at ith band.

Figure 4. Hamming Window applied to each frame-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Use of speech spectrum for modifying work domain on signals from time to frequency is made possible using Fourier coefficients. At such applications the rapid and practical way of estimating the spectrum is use of rapid Fourier changes.

Figure 3 Frame blocking of the speech signal-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

The next frame will begin M samples (i.e. 156 samples) after the first frame, and it will overlap the first frame by N-M samples (256 – 156 = 100 samples). Then the third frame will start at 2M samples after the first frame and it will overlap first frame by N-2M. The fourth frame will start at 3M samples after the first, and it will overlap it by N-3M. The process will continue until all input signal is accounted for. The result of this step plotted using MATLAB plot command and displayed in Figure 3. Figure 3 Frame blocking of the speech signal The next step in the processing is to window each individual frame so as to minimize the signal discontinuities at the beginning and end of each frame. The concept here is to minimize the spectral distortion by using the window to taper the signal to zero at the beginning and end of each frame. If we define the window as w(n), 0 ≤ n ≤ N −1, where N is the number of samples in each frame, then the result of windowing is the signal y (n) = x (n)w(n), 0 ≤ n ≤ N −1 l l

opencc-by-4.0Dec 2010View details →

Figure 1. The Structure of the Automatic Translate Voice to Sign Language Animation System-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

All technologies of voice recognition, speaker identification and verification, each has its own advantages and disadvantages and may requires different treatments and techniques. The choice of which technology to use is application-specific. At the highest level, all voice recognition systems contain two main modules: feature extraction and feature matching. Feature extraction is the process that extracts a small amount of data from the voice signal that can later be used to represent each word. Feature matching involves the actual procedure to identify the unknown word by comparing extracted features from his/her voice input with the ones from a set of known words. A wide range of possibilities exist for parametrically representing the speech signal for the voice recognition task, such as Linear Prediction Coding (LPC), RASTA-PLP and Mel-Frequency Cepstrum Coefficients (MFCC).

Figure 8. Speech signal after frequency wrapping-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Use of speech spectrum for changing the work domain on the signal from time to frequency is used via Fourier conventions. The rapid and practical way of estimating Spectrum in such applications is employment of rapid Fourier transform. The final stage is extracting particularities is use of discrete cosine to return particularities to the time domain and converse FFT approximation. Major advantage of this method is decrease of number of particularities of number of filter from f N to c N in which c f N ≤ N . In addition, doing so, includes making independent of the obtained particularity and rendering them non dependant which leads matrix covariance features to become axial. The following equation shows this point.

Figure 7. Speech signal after FFT-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Figure 2. MFCC Block Diagram Step 1-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Mel Frequency Cepstral Coefficients (MFCC) are coefficients that represent audio, based on perception. It is derived from the Fourier Transform (FFT) or the Discrete Cosine Transform (DCT) of the audio clip. The basic difference between the FFT/DCT and the MFCC is that in the MFCC, the frequency bands are positioned logarithmically (on the Mel scale) which approximates the human auditory system's response more closely than the linearly spaced frequency bands of FFT or DCT. This allows for better processing of data. The main purpose of the MFCC processor is to mimic the behavior of the human ears. Overall the MFCC process has 5 steps that show in figure 2.

Figure 11. A sample of speech to facial animation system-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

To increase the autonomy of deaf and hard of hearing people in their day-to-day professional and social lives, in this paper design and initial implementation of a new approach based on MFCC and Vector Quantization Method is described. This approach includes analyses of speech to animate the talking head. Our future work will include the conception of new test types and performance patterns. We are particularly interested in extending this approach to testing to include implementation of applications under real-time constraints

Figure 5. Mel-spaced filter bank-Development an Automatic Speech to Facial Animation Conversion for Improve Deaf Lives

Physiologic changes show that human comprehension of frequency content of sound does not obey a linear space. Therefore for each individual it is computed and measure with a real frequency of sound peak at Mel seal. Using the below equation one can change the frequency at Hertz scale to Mel Scale.

opencc-by-4.0May 2024View details →

Attempted Speech on Single Words [intracortical array][T16][BG2]

<h2>Summary of the Data and Experiment</h2> <h3>General Overview</h3> The data provided come from an attempted speech experiment involving a participant (T16) in the BrainGate2 clinical trial. The dataset includes neural recordings from four microelectrode arrays implanted in the left precentral gyrus. The primary focus is on the array located in the ventral precentral gyrus (6v), which is associated with speech-related activity. <h3>Data Description</h3> The neural data consists of two main streams: <ol> <li>Log Local Field Potential (LFP) Power: <ul> <li>Description: Extracellular neural data from the lower frequency bands.</li> <li>Sampling Rate: Preprocessed to 2 kHz.</li> <li>Filtering: Bandpass filtered with a 150-450 Hz band.</li> <li>Scale: Log-scale</li> <li>Bin size: 20ms Summed</li> <li>Shape: 128457x64 (time by channels).</li> </ul> </li> <li>Binned Threshold Crossings: <ul> <li>Description: Extracellularly collected neural spiking activity related to attempted speech.</li> <li>Threshold: -4.5 RMS</li> <li>Bin size: 20ms Summed</li> <li>Shape: 128457x64 (time by channels)</li> </ul> </li> </ol> The included data are within a single mat file with the keys `trial_info_df`, `lfp_matrix`, `spikes_matrix`, and `timevector`. <h3>Trial Information</h3> The `trial_info_df` key contains information about each trial in the form of a pandas DataFrame with the following columns: <ul> <li>block_num</li> <li>cue</li> <li>start_time</li> <li>go_cue_time</li> <li>end_time</li> </ul> <h3>Participant and Experiment Information</h3> <ul> <li>Participant T16: Right-handed, 52-year-old woman with tetraplegia and dysarthria due to a pontine stroke.</li> <li>Implantation: Four 64-channel intracortical microelectrode arrays in the left precentral gyrus.</li> <li>Recording Day: Data collected 69 days post-implantation.</li> <li>Recording Platform: Backend for Realtime Asynchronous Neural Decoding (BRAND) platform.</li> <li>Spiking Data Extraction: Linear regression referencing, bandpass filtering (250-5000 Hz), and threshold crossings identification (-4.5 RMS threshold).</li> <li>LFP power Extraction: Band pass filtering (150 - 450 Hz), and notch filtering at harmonics of 60 Hz.</li> </ul> <h3>Experimental Task</h3> <ul> <li>Task: Cued speech task where Participant T16 vocalized words presented on a screen.</li> <li>Word Bank: Compiled from a 50-word vocabulary by Moses et al.</li> <li>Trial Structure: <ul> <li>A red square appeared below a word.</li> <li>After 1500 ms, the square turned green, cueing the participant to vocalize the word.</li> <li>The trial ended after the participant finished speaking, with a 1000 ms interval before the next trial.</li> </ul> </li> </ul> <h2>Loading and Handling the Data</h2> The data can be loaded into Python using `scipy.io.loadmat`: <h3>Loading the Data</h3> <blockquote> import scipy.io import pandas as pd # Load the .mat file data = scipy.io.loadmat('path_to_mat_file.mat') # Extracting the trial information trial_info_df = pd.DataFrame(data['trial_info_df']) trial_info_df.columns = ['block_num', 'cue', 'start_time', 'go_cue_time', 'end_time'] # Extracting neural data matrices lfp_matrix = data['lfp_matrix'] spikes_matrix = data['spikes_matrix'] # Extracting time vector timevector = data['timevector'] </blockquote> <h3>Wrapping Trial Information in a DataFrame</h3> <blockquote> trial_info_df = pd.DataFrame(data['trial_info_df'], columns=['block_num', 'cue', 'start_time', 'go_cue_time', 'end_time']) </blockquote> <h3>Exploring the Data</h3> <ul> <li>Log Local Field Potential (LFP) Power: `lfp_matrix`</li> <li>Threshold Crossings (Spikes): `spikes_matrix`</li> <li>Time Vector: `timevector`</li> </ul> <h2>Support</h2> This work was supported by NIH-NINDS/OD DP2NS127291 (CP), NIH-NICHD F32HD112173 (SRN), NIH K08NS060223 (MS), R01NS112942 (MS), and RF1NS125026 (MS). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, or the Department of Veterans Affairs, or the United States Government. *Caution: Investigational Device. Limited by federal law to investigational use.

opencc-by-sa-4.0Jun 2024View details →

AnglistikVoices: L2 English speech dataset

<h1>AnglistikVoices: an L2 English speech dataset </h1> This repository contains an L2 (second language) English speech corpus consisting of 74 minutes of recorded audio from 15 non-native English speaking participants. The dataset was created as part of a university course, with all participants being students who are also the authors of this dataset. <h2>Dataset Specifications</h2> <ul> <li>Total participants: 15 non-native English speakers</li> <li>Total audio duration: 74 minutes</li> <li>Recordings per participant: 60 audio samples each</li> <li>Sentence alignment: Available for 8 out of 15 participant</li> <li>Recording equipment: Audio-Technica ATM75 microphone</li> <li>Stimuli: All sentences are from the Artie Bias Corpus (https://github.com/artie-inc/artie-bias-corpus)</li> <li>Recording environment: Recording booth</li> </ul> The dataset contains individual recordings of non-native English speakers organized by participant ID. For 8 participants, sentence-level alignments are provided. All recordings were captured in a controlled acoustic environment using Audio-Technica ATM75 microphone to ensure high audio quality.  The recordings consist of spoken English utterances from each participant. Detailed linguistic profiles for each participant are available in the metadata.xlsx file, which is indexed by participant ID and contains information on native language, proficiency level, language learning history, and other relevant linguistic background data. The audio files are organized by participant ID, matching the identifiers used in the metadata file for easy cross-referencing between the audio recordings and participant linguistic profiles. <h2>Authors and Contributors</h2> This dataset was created by the student participants themselves as part of their coursework. Course Instructor: Akhilesh Kakolu Ramarao Teaching Assistant: Anna Sophia Stein If you have any questions, you can contact: kakolura@hhu.de If you use this dataset in your research, please cite: <pre><code>@dataset{kakolu_ramarao_2024_anglistikvoices, author = {Kakolu Ramarao, A. and Stein, A. S. and Tahiri, A. and Rodrigues, D. C. and Antonia Weismann, C. and Schäfer, O. S. and Kaczor, J. and Tran, N. H. and Elena Telaar, C. and Bauer, L. and Jütten, M. and Mafuta, C. and Agelopoulou, V. V. and Grabowski, Q. A. G.}, title = {AnglistikVoices: L2 English speech dataset}, publisher = {Zenodo}, version = {v1.0.0}, year = {2024}, month = jun, doi = {10.5281/zenodo.12525952}, url = {https://doi.org/10.5281/zenodo.12525952}, note = {LabPhon 19, Hanyang Institute for Phonetics and Cognitive Sciences of Language (HIPCS), Hanyang University in Seoul, Korea} }</code></pre>