Live transcription and intelligence

Live transcription processes a conversation while it is happening. Audio streams over WebSocket, transcript segments arrive as they are recognized, and optional intelligence features turn the running conversation into events that agents and supervisors can act on.

Live capabilities are account-controlled. If a required capability is unavailable, contact your Speechlytics representative.

Session lifecycle

POST /api/v1/live-speech/start
        ↓
connect to the returned WebSocket URL
        ↓
stream PCM audio and receive transcription messages
        ↓
POST /api/v1/live-speech/stop
        ↓
recording and transcript enter the post-call pipeline

Both REST operations require a JWT. Starting a session returns its ID, WebSocket URL, and viewer PIN. See Start a live transcription for request examples and streaming requirements.

Mono and stereo

	Stereo	Mono
Agent	Channel 1, returned as Left	Shared Both channel
Customer	Channel 0, returned as Right	Shared Both channel
Per-speaker metrics	Available	Not separated
Per-speaker translation	Available	Single shared target

Choose the channel count that matches the audio source. Speechlytics does not estimate speaker-specific values when mono audio does not preserve speaker separation.

Optional live capabilities

Keyword detection and alerts

Live keyword detection finds configured phrases as finalized segments arrive. Alert rules can turn detections, detection counts, or negative-sentiment thresholds into informational, warning, or critical events.

Live keyword detection is independent of post-call keyword matching: one operates during the conversation and the other after the transcript is complete.

Conversation metrics

Metrics are calculated from the live transcript and audio duration. They include talk time, silence, interruptions, longest monologue, response time, speaking rate, questions, unanswered questions, and filler words.

Mono conversations do not report speaker-separated metrics, and unanswered questions remain zero because there is no reliable second channel to inspect.

Sentiment, intent, and conversation state

Live classification can report customer sentiment, frustration, and call intent. Conversation State adds a running structured view of the stage, issues, entities, requested actions, and resolved or unresolved items.

Conversation State depends on live sentiment classification. Summary and coaching use that state as context.

Summary and agent coaching

Live Summary maintains a concise account of the conversation and the customer's current request. Agent Coaching can add warnings, a suggested response, next best action, recommended questions, opportunities, and compliance reminders.

Coaching is agent-facing and is not returned to a public PIN viewer.

Language detection and translation

Live Language Detection records the language detected for incoming segments. A session can use a fixed translation target, re-detect language for every segment, or automatically translate each speaker toward the language detected for the other speaker.

Until both sides have a detected language, automatic direction falls back to the configured translation target.

Consuming live events

When the Events API is available, events can be read by cursor:

GET /api/v1/calls/{id}/live-events?afterId=0&limit=200

Store the largest returned event ID and send it as the next afterId. This makes reconnection safe and avoids rereading the full event history.

The same event stream is available over /ws/live-events?callId={id} for low-latency consumers. The REST endpoint is simpler to recover after a network interruption.

When a live call ends

Stopping the session finalizes the stored call. Enabled post-call features can then perform redaction, keyword matching, translation, summaries, analytics, and custom-question grading. Retrieve the completed result with GET /api/v1/calls/{id}/status.

Troubleshooting

Symptom	What to check
Start returns 403	Live transcription is not available for the account.
No recognition messages	Confirm authentication, PCM format, sample rate, channel count, and real-time pacing.
Speakers are reversed	Channel 1 must contain the agent and channel 0 the customer.
Live events return 403	The Events API is not available for the account.
A panel is missing	Its capability or prerequisite is not available for the account.
Translation starts with the fallback target	Automatic direction has not detected both speakers' languages yet.