Syarotto/speech-to-text-translation

Code for the final project "Speech-to-Text Translation in Swahili" of LING 575C: Speech Technology for Endangered Languages # speech-to-text-translation Speech-to-text translation for Swahili ## Installation This system is tested on `python==3.8.0`, and a list of required packages can be found at requirements.txt. It is recommended to use Anaconda or Miniconda to set up the environment. With `conda` installed, you may run: ``` conda create -n 575c python==3.8.0 conda activate 575c pip install -r requirements.txt conda install -c conda-forge libsndfile ``` The endangered language parallel dataset LORELEI can be accessed on the cluster by: ``` newgrp lorelei cd /corpora/LORELEI/LDC2017E64_LORELEI_Swahili_Representative_Language_Pack_Translation_Annotation_Grammar_Lexicon_and_Tools_V1.0 ``` To set up the Google Cloud Speech service, follow the instructions on Install the Google Cloud CLI and Setting up a Python development environment, then install `google-cloud-speech` through `pip`. You also need to generate the credential key given Getting started with authentication, where the service account can be found in this link. Name the credential json file as `credential.json` and place it under the root directory of this project. ## Resources: - End-to-End_Speech-to-Text_Translation - Huggingface/speech-to-text - fairseq/speech-to-text - alokmatta/wav2vec2-large-xlsr-53-sw - MUST-C - facebook/m2m100_418M - IWSLT 2021 Swahili - Google Cloud Speech