Features

Complete feature reference for Spoke Work.

Real-time Transcription

Live speech-to-text powered by Deepgram WebSocket streaming:

• 16 kHz PCM audio streamed in real-time
• Speaker diarization with automatic speaker separation
• Auto-punctuation and paragraph formatting
• Auto-reconnect on WebSocket failure (up to 3 attempts)

Real-time Translation

Streaming translation via Supabase Edge Functions:

• 6 target languages: Chinese, Spanish, Japanese, Korean, French, German
• Context-aware: Uses recent transcript segments for better translations
• Server-Sent Events delivery for low-latency streaming
• Smart debounce: 800ms delay with 40+ character threshold

Channel Management

Organize meetings by channel:

• Meeting types: In-person, Zoom, Google Meet, Microsoft Teams
• Member management: Add participants with names and role descriptions
• Intent tracking: Link meeting objectives to channels
• Recurring sessions: Reuse channel configuration across meetings

AI Speaker Inference

Maps anonymous speaker labels to real participant names:

• Uses participant names and descriptions from channel configuration
• Context-driven inference from conversation content
• Can restrict identification to known participants only
• Runs with 30-second timeout and abort support

Audio Management

• WAV recording with 16-bit PCM at 16 kHz
• Upload with retry: 3 attempts with exponential backoff
• Audio level visualization during recording
• Keep-awake: Screen stays on during active recording
• 5-second replay buffer for reviewing recent audio

Insights and Intents

After recording:

• Intents: Define meeting objectives and track outcomes
• Insights: AI-extracted action items, summaries, and key points
• Custom analysis: Run custom prompts against the transcript