This is a speech-to-text experiment consisting in transcribing episodes of a podcast and making the transcriptions searchable through a web application.
Steps involved:
- Retrieve podcast episodes
- Cut initial and ending credits
- Upload cut episodes to Azure Storage
- Transcribe episodes in Azure Storage using Azure Speech Services
- Expose transcriptions as a search engine web application
- Install required Python packages
python -m venv venv source venv/bin/activate python -m pip install -r requirements.txt - Setup secrets in
.envfileAZURE_SUB_KEY: Azure Speech Services subscription key AZURE_REGION: Azure Speech Services region AZURE_STORAGE_CONNECTION_STRING: Azure Storage Account connection key (Storage account > Access keys) AZURE_BLOB_CONTAINER_URI: Azure Container URL (Container > Shared access token - Read and List rights) ILPOST_USER: Il Post username ILPOST_PASS: Il Post password - Setup output folders
mkdir -p output/episodes-original output/episodes/cut output/transcriptions ln -s $(pwd)/output/transcriptions api/transcriptions
Run ./create_transcriptions.sh.
Start frontend development server:
cd frontend
npm start
# to build the React application in ./api/static
npm run buildStart API server:
cd api
flask --debug --app server run