Blvisse/Speech-Recognition

This project aims at creating a speech-to-text model for the Swahili language. The model converts audio Wav files into melSpectograms which are then fed into a deep learning model which then transcribes to the predicted text # Speech-Recognition ## Introduction The project is inspired by the use of speech recognition systems in everyday life . Speech recognition systems sych as Siri, Alexa and google assitant convert speaach into action and hence automate and make work easier. This speech recognition models are however limited to a select few languages creating a huge langauge barrier. The project aims to integrate swahili langauge into a speech to text system that allows the user to echo in commands in swahili which will then be converted into text. The speach-to-text system will be integrated in a mobile app for food purchase. The project seeks to use deep learning models that is capable of transcribing speech to text, with the aim of making the model accurate and robust against background noise ## Data The data collected and reference for the swahili dataset can be found below: ### PUBLICATION ON SWAHILI SPEECH & LM DATA More details on the corpus and how it was collected can be found on the following publication (please cite this bibtex if you use this data) @InProceedings { gelas:hal-00954048, author = {Gelas, Hadrien and Besacier, Laurent and Pellegrino, Francois}, title = {{D}evelopments of {S}wahili resources for an automatic speech recognition system}, booktitle = {{SLTU} - {W}orkshop on {S}poken {L}anguage {T}echnologies for {U}nder-{R}esourced {L}anguages}, year = {2012}, address = {Cape-Town, Afrique Du Sud}, abstract = {no abstract}, x-international-audience = {yes}, url = {hal.inria.fr, } ### SWAHILI SPEECH CORPUS Directory: /data/train Files: text (training transcription), wav.scp (file id and path), utt2spk (file id and audio id), spk2utt (audio id and file id), wav (.wav files). For more information about the format, please refer to Kaldi website kaldi-asr.org Description: training data in Kaldi format about 10 hours. Note: The path of wav files in wav.scp have to be modified to point to the actual locatiion. Directo …