CMU Sphinx

Mentors: Student: Mritunjay Goutam

<p>As mention in the title, it is a 3D animation model of the motor control of speech production. Our long range goal is to implement the model completely and test it exhaustively, in the belief that doing so will significantly advance our understanding of speech motor control. The model is designed to have properties that are as faithful as possible to data from speakers– including measures of brain function, speech motor control mechanisms,physiolog, an atom, biomechanics, and acoustics.</p>

Mentors: Student: Rishi Rajasekaran

<p>The aim of this project is to write support code for obtaining accurate transcriptions, exemplar pronunciations, and phonetic and part-of-speech labeling.</p> <p>One of the important concerns in Natural Language Processing is the availability of gold standard information such as transcripts of speech, part-of-speech tagged sentences, word pronunciation examples etc., which aid in training and evaluating the performance of speech and text processing algorithms. This requires extensive collection, processing and validation of information from a variety of sources.</p> <p>The data needs to be normalized across these sources, evaluated based on quality, quantity and other metrics. This requires a structured system which allows for the data collection and management tasks to be done effectively. This is especially important to CMU Sphinx contributors who require such data for their work, such as training and evaluating their speech and text models.</p> <p>The project involves the integration of a Flask and SQLite based leaderboard web application with a transcription record system into a broader data aggregation and processing framework for the aforementioned tasks.</p>

Mentors: Student: Vishal Agrawal

<p>Automatic speech recognition systems are traditionally based on Hidden Markov models(HMMs) as it had given the best results in the past. However due to recent advances in neural networks, recurrent neural network(RNN) models have became competitive and surpassed HMMs in speech recognition task. RNN models also have the advantage of being used in an end-to-end model and don't require extensive hand engineering of features like HMM. In this project I aim to implement a LSTM(Long Short Term Memory) based model as based on state of the art research and make it easily available using PocketSphinx like API.</p>

Mentors: Student: Hammad Abdullah

<p>CMUsphinx is a fast and flexible open source speech recognition toolkit. Currently it uses GMM acoustic models and it gives reasonable accuracy. This project is geared towards incorporating a state of the art MLP-HMM model into sphinx, substantially decreasing the error rates. A ResNet with MLPs instead of convolution networks will be trained and tested on TEDLIUM corpus.The aim is to achieve best-in-category accuracy in recognition of Ted talks.This project will bring CMUsphinx at par with other powerful speech toolkits like KALDI and will help Sphinx users worldwide get better accuracy in their ASR projects.</p>

Mentors: Student: Pavel Denisov

<p>The goal of this project is to extend PocketSphinx so that it can calculate and provide alignments and acoustic scores for diphones similarly to how it already does for phonemes. This will help with learner analytics and adaptive instruction for pronunciation evaluation, assessment, and remediation, and can increase speech recognition accuracy and speech recognition system generality. Moreover, diphones may be helpful for new acoustic models, and can inform the construction of different kinds of language models.</p>