{
  "fetched_at": "2026-10-19T16:35:08Z",
  "source_url": "https://www.kaggle.com/competitions/11-785-homework-1-part-2-fall-18-late/data",
  "pages": [
    {
      "name": "rules",
      "content": "\r\n\t<!-- Begin Base Rules -->\r\n    <h3>One account per participant</h3>\r\n    <p>You cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.</p>\r\n    <h3>No private sharing outside teams</h3>\r\n    <p>\r\n        Privately sharing code or data outside of teams is not permitted.\r\n        It's okay to share code if made available to all participants on the forums.\r\n    </p>\r\n    <h3>Team Mergers</h3>\r\n    <p>Team mergers are not allowed in this competition.</p>\r\n\r\n    <h3>Team Limits</h3>\r\n    <p>There is no maximum team size.</p>\r\n    <h3>Submission Limits</h3>\r\n    <p>You may submit a maximum of 2 entries per day.</p>\r\n    <p>You may select up to 2 final submissions for judging.</p>\r\n    <h3>Competition Timeline</h3>\r\n    <p>Start Date: <strong>TBA</strong></p>\r\n    <p>Merger Deadline: <strong>None</strong></p>\r\n    <p>Entry Deadline: <strong>None</strong></p>\r\n    <p>End Date: <strong>2/1/2014 12:00 AM UTC</strong></p>\r\n\t<!-- End Base Rules -->\r\n<p>The rules page is a bit special: it will auto-update the text based on the parameters of the competition. The rules you need to add here are any additional rules relating to dataset usage. You may also put grading information here.</p>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "data-description",
      "content": "Please find data in previous links.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "<h1>Frame Level Classification of Speech</h1>\n<p>In this challenge you will take your knowledge of feedforward neural networks and apply it to a more useful task than recognizing handwritten digits: speech recognition. You are provided a dataset of audio recordings (utterances) and their phoneme state (subphoneme) labels. The data comes from articles published in the Wall Street Journal (WSJ) that are read aloud and labelled using the original text. If you have not encountered speech data before or have not heard of phonemes or spectrograms, we will clarify the problem further.</p>\n<b>The training data comprises of:</b>\n<ol>\n<li>Speech recordings (raw mel spectrogram frames)</li>\n<li>Frame-level phoneme state labels</li>\n</ol>\n<b>The test data comprises of:</b>\n<ol><li>Speech recordings (raw mel spectrogram frames)</li>\n<li>Phoneme state labels are not given</li>\n</ol>\n<p>Your job is to identify the phoneme state label for each frame in the test data set. It is important to note that utterances are of variable length. We are providing you code to load and parse the raw files into the expected format. You can find that code <a href=\"https://gist.github.com/rbrigden/d611ef4f2eccddb5581e0ac617ce38eb\">here</a>.</p>\n<h2>Phonemes and Phoneme States</h2>\n<p>As letters are the atomic elements of written language, phonemes are the atomic elements of speech. It is crucial for us to have a means to distinguish different sounds in speech that may or may not represent the same letter or combinations of letters in the written alphabet.</p>\n<p>For this challenge we will consider 46 phonemes in the english language.</p>\n<p>[\"+BREATH+\", \"+COUGH+\", \"+NOISE+\", \"+SMACK+\", \"+UH+\", \"+UM+\", \"AA\", \"AE\", \"AH\", \"AO\", \"AW\", \"AY\", \"B\", \"CH\", \"D\", \"DH\", \"EH\", \"ER\", \"EY\", \"F\", \"G\", \"HH\", \"IH\", \"IY\", \"JH\", \"K\", \"L\", \"M\", \"N\", \"NG\", \"OW\", \"OY\", \"P\", \"R\", \"S\", \"SH\", \"SIL\", \"T\", \"TH\", \"UH\", \"UW\", \"V\", \"W\", \"Y\", \"Z\", \"ZH\"]</p>\n<p>A powerful technique in speech recognition is to model speech as a markov process with unobserved states. This model considers observed speech to be dependent on unobserved state transitions. We refer to these unobserved states as phoneme states or subphonemes. For each phoneme, there are 3 respective phoneme states. Therefore for our 46 phonemes, there exist 138 respective phoneme states. The transition graph of the phoneme states for a given phoneme is as follows:</p>\n<img src=\"https://github.com/Nihar94/Assignment_Store/blob/master/pstates.jpeg?raw=true\">\n<p>Hidden Markov Models (HMMs) estimate the parameters of this unobserved markov process (transition and emission probabilities) that maximize the likelyhood of the observed speech data. Your task is to instead take a model-free approach and classify mel spectrogram frames using a neural network that takes a frame (plus optional context) and outputs class probabilities for all 138 phoneme states. Performance on the task will be measured by classification accuracy on a held out set of labelled mel spectrogram frames. Training/dev labels are provided as integers [0-137].</p>\n<h2>Speech Representation</h2>\n<p>As a first step, the speech (audio file) must be converted into a feature representation (matrix form) that can be fed into the network.</p>\n<p>In our representation, utterances have been converted to \"mel-spectrograms\" (you do not have to convert anything, just use the given mel-spectrograms), which are pictorial representations that characterize how the frequency content of the signal varies with time. The frequency domain of the audio signal provides more useful features for distinguishing phonemes.</p>\n<p>For a more intuitive understanding, consider attempting to determine which instruments are playing in an orchestra given an audio recording of a performance. By looking only at the amplitude of the signal of the orchestra over time, it is nearly impossible to distinguish one source from another. But if the signal is transformed into the frequency domain, we can use our knowledge that flutes produce higher frequency sounds and bassoons produce lower frequency sounds. In speech, a similar phenomenon is observed when the vocal tract produces sounds at varying frequencies.</p>\n<p>To convert the speech to a mel-spectrogram, it is segmented into little \"frames\", each 25ms wide, where the \"stride\" between adjacent frames is 10ms. Thus we get 100 such frames per second of speech.</p>\n<p>From each frame, we compute a single \"mel spectral\" vector, where the components of the vector represent the (log) energy in the signal in different frequency bands. In the data we have given you, we have 40-dimensional mel-spectral vectors, i.e. we have computed energies in 40 frequency bands.</p>\n<p>Thus, we get 100 40-dimensional mel spectral (row) vectors per second of speech in the recording. Each one of these vectors is referred to as a frame. The details of how mel spectrograms are computed from speech is explained <a href=\"http://haythamfayek.com/2016/04/21/speech-processing-for-machine-learning.html\">here</a>.</p>\n<p>Thus, for a T-second recording, the entire spectrogram is a 100*T x 40 matrix, comprising 100*T 40- dimensional vectors (at 100 vectors (frames) per second).</p>\n<h2>Data Files</h2>\n<p>You can find the training and test data under the 'Data' tab</p>\n<h3>Optional Readings</h3>\n<p>Read the Model Architecture section for some tips and advice on building your classifier.</p>\n<p>Go through <a href=\"http://haythamfayek.com/2016/04/21/speech-processing-for-machine-learning.html\">this link</a> to understand how Mel-Spectrograms are generated</p> ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "<h2>Evaluation Metric</h2>\n<p>The evaluation metric for this competition is frame-level accuracy. There are a total of 169,656 frames in the test set. You will be ranked by unweighted accuracy on those phoneme state labels.</p>\n<h2>Submission Format</h2>\n<p>Submission files should contain two columns: Id and Label.</p>\n<ul>\n<li><b>Id</b>: the 0-based index of the frame in the test set [0-169,655]</li>\n<li><b>Label</b>: the predicted label of the phoneme [0-137]</li>\n<p>Id=0 is the first frame in the first utterance. Id=169,656 is the last frame in the last utterance.\nA sample submission file is available on the Data page.</p>\n<h2>Rules</h2>\n<p>Your model <b>must</b> use an MLP (no RNNs or ConvNets). We will check your code submitted to Autolab to make sure this is the case. However, you are free to use auto-differentiation libraries like PyTorch or Tensorflow.</p>\n<p> Note: Support will only be provided for Pytorch.</p>\n<h2>Autolab</h2>\n<p><b>Important</b>: In order to receive a grade for this portion of the assignment, you must turn in all of the code you used to generate your Kaggle submission on Autolab. This means that <b>all of your code for HW1pt2 and a file called 'kaggle.txt' containing only your Kaggle user ID</b> should be put in a tarball (.tar file) and submitted here. Autolab will not be autograding your code, this is only to confirm that you wrote the code for the assignment yourself.\n<h2>Grading</h2>\n<ul>\n<li>Submissions that outperform random classification are guaranteed a C grade.</li>\n<li>Submissions that outperform a baseline provided by the course staff are guarenteed a B grade.</li>\n<li>Top scoring (lowest error) submissions will receive grades in the A range.</li>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Implementation Tips",
      "content": "<h1>Tips and Implementation Guidance</h1>\n<h2>Network Architecture</h2>\n<p>Feel free to experiment with different architectures. The simplest approach is to build a feedforward network that takes in a single frame (input size 40) and outputs probabilities across phoneme states using a softmax output to normalize the logits.</p>\n<h2>Context</h2>\n<p>Temporal context is important for distinguishing elements of speech and you can experiment with adding context to your ANN model. For example, you can try concatenating k mel spectrogram frames around the current time step. This technique would make the size of the input vector 40 * (2k + 1). You will need to adjust your data preprocessing and batching logic to adapt for frame context. Writing code that will let you experiment with variable context sizes easily is recommended.</p>\n<h2>Data Batching</h2>\n<p>It is important to consider that the data is provided in the form of utterance samples and not frame samples. The property of mini-batch SGD that gradients over all batches in an epoch approximate the true gradient only holds if each batch is IID (Independent and Identically Distributed). If we sample uniformly over utterances and also sample uniformly over frames within the utterance, our batches will surely not be IID. For example, different utterances will have considerably different numbers of phoneme instances. Sampling uniformly biases the training data by over-representing phonemes in the utterance with fewer phonemes (and therefore its phoneme states). You can try a different sampling technique to batch your training data to mitigate this issue.</p>\n\n<p>Many approaches are possible. See what can get you to the top of the leaderboard!</p>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}