Skip to content

how-it-works.md

Latest commit

 

History

History
81 lines (54 loc) · 2.34 KB

how-it-works.md

File metadata and controls

81 lines (54 loc) · 2.34 KB

How It Works

The Audio Analyzer microservice provides capability to transcribe audio from input video file.

Architecture Overview

High-Level Architecture Diagram

Architecture Diagram Figure 1: High-level system view demonstrating the microservice.

Inputs

  • Video file: Provides reference to the video file to be used as input. The file can be ingested from file system or from MinIO store.
  • API Requests: REST API calls for operations such as Whisper model selection, Health, and generating transcription.

Processing Pipeline

  • Transcription: The transcription process involves demuxing to get the audio channel in the input video file and providing the same to the Whisper model for transcription. The transcription process follows the well documented Whisper model internals.

Outputs

  • API Responses: Provides feedback on operations, including success or error messages, and returns output for the requested API call.

API Endpoints

Health Check

GET /api/v1/health

Returns the status of the API service.

Speech Transcription

POST /api/v1/transcriptions

Transcribes speech from a video file using either direct upload or a MinIO source.

Request Parameters (Two ways to provide video source):

Option 1: Direct File Upload

  • file: The video file to transcribe

Option 2: MinIO Source

  • minio_bucket: Name of the MinIO bucket where the video is stored
  • video_id: ID/prefix of the video in the bucket (optional)
  • video_name: Name of the video file in the bucket

Common Parameters:

  • include_timestamps (optional): Whether to include timestamps in the output (default: true)
  • device (optional): Compute device to use - 'cpu', 'gpu', or 'auto' (default: from settings)
  • model_name (optional): Name of the whisper model to use (default: from settings)
  • language (optional): Language code for transcription

Example Response:

{
  "status": "completed",
  "message": "Transcription completed successfully",
  "job_id": "1234-5678-90ab-cdef",
  "transcript_path": "minio://video_19/audio_description/1234-5678.srt",
  "video_name": "example.mp4",
  "video_duration": 120.5
}

Supporting Resources