Spoke Class

Recording & Processing

How Spoke Class records lectures and processes them through the AI extraction pipeline.

Recording

Audio capture uses the same 16 kHz mono PCM pipeline as all Spoke apps, streamed to Deepgram via WebSocket for real-time transcription.

Key differences from Spoke Work:

  • 12 source languages supported (vs. English-only in Spoke Work)
  • Speaker diarization enabled for identifying professor vs. students
  • Translation streams alongside the original transcript in real-time

Processing Pipeline

After recording stops, the audio enters a background processing pipeline:

  1. Upload — Audio file is uploaded to Supabase Storage
  2. Transcription — Full transcript is generated with speaker labels
  3. Extraction — AI runs each linked intent against the transcript
  4. Completion — Results are stored and available for review

Status Tracking

Each recording shows its processing status:

StatusMeaning
PendingQueued for processing
ProcessingActively being transcribed/extracted
CompletedAll processing finished successfully
FailedAn error occurred (can be retried)

A progress percentage is displayed during the processing stage. Failed recordings can be retried from the recording detail screen.

Viewing Results

After processing completes:

  • Transcript — Full transcript with speaker labels and timestamps
  • Translations — Translated text if translation was enabled
  • Extracted Content — Results from each linked intent, displayed as structured sections
  • Playback — Listen to the recording with synchronized transcript highlighting

On this page