How sundar0989/speech-to-text works
Project type and purpose
• This is a small open-source demonstration/toolset implemented as Jupyter notebooks. Its sole purpose is to convert long-duration audio files into text using two cloud-provider speech-to-text APIs (AWS Transcribe and Google LongAudio API), with optional speaker diarization.
File structure
• Top level:
• Four notebooks, each implementing one variant of the pipeline:
– AWS_transcribe.ipynb
– AWS_transcribe_with_Speaker_diarization.ipynb
– Google_Longaudio_API_without_speaker_diarization.ipynb
– Google_Longaudio_API_speaker_diarization.ipynb
• README.md describing the overall capability.
• No separate “src” or “scripts” folder—each notebook contains both orchestration logic and SDK calls.
README for additional insights
• Confirms that this is an asynchronous, long-audio speech-to-text solution.
• Mentions optional speaker diarization.
a. Main components
1. User / Data Scientist environment (Jupyter Lab / Notebook host)
2. Notebook orchestration code
3. Cloud SDKs (AWS SDK v2/v3 or Google Cloud client libraries)
4. Cloud speech-to-text service endpoints:
• AWS Transcribe
• Google LongAudio API
5. Audio storage (could be local file system or S3/GCS)
6. Transcription result store (local files, in-notebook variables, or cloud storage)
b. Relationships & interactions
• The user uploads or references a long-audio file in the notebook.
• Notebook invokes cloud SDK → submits transcription job asynchronously.
• Cloud service polls or receives events via SDK until transcription completes.
• Result is fetched back into the notebook.
• Optionally, speaker-diarization metadata is retrieved and post-processed.
c. Architectural patterns & design principles
• “Pipeline” pattern: sequential stages (submit job → wait → fetch result → process).
• Asynchronous job orchestration using polling or callbacks.
• Separation of concerns: notebook as driver, cloud as compute.
d. Key technologies & libraries
• Python (within Jupyter)
• Boto3 or AWS SDK for Python
• google-cloud-speech or LongAudio client library
• Optional: pandas or JSON for result parsing
Step-by-step approach for the principal software engineer
2. Inside it, show notebooks calling two modules: AWS_Module and Google_Module.
4. Under each cloud service, show “S3/GCS Storage” if used.