Signals

A Signal turns recorded conversations into a table. You describe who takes part, what kinds of conversation to expect and what you want to know; Tavnit transcribes the audio, works out who said what, splits the recording into separate conversations and fills your columns turn by turn.

The Signal is the configuration. Each audio file you run through it produces a Wave: one execution with its own transcript, player and output table. A shop-floor recording of a whole shift can hold dozens of sales conversations; a Wave finds each one and structures it separately.

• Checking whether salespeople mention a promotion, and which products customers ask about
• Scoring support or intake calls against a checklist of things that should happen
• Turning interviews into rows you can filter, chart and store in a Bucket

Only a name and at least one member type are required. Everything else is optional and adds columns or filters to the output.

Descriptions matter: the AI reads them to decide who is who and when a rule or category applies. Names must be unique within each list and can't reuse a built-in column name.

1. 1Open “Signals” in the sidebar and click “Create Signal”.
2. 2Choose how to start: “Conversation preset” (a sales, support or interview setup), “From an existing signal” (duplicates its full configuration) or “Start from scratch”.
3. 3Enter a Signal Name (at least 3 characters) and define at least one member type.
4. 4Optionally add interaction types, per-turn rules, extraction fields and content categories, then conversation-level rules and category groups.
5. 5Under Options, decide whether to exclude unmatched interactions, add a Wave ID column or send results to a webhook.
6. 6Click “Create Signal”. Recorders, Bucket export and the Signal ID for the API are set up on the Signal's page once it exists.

To change the configuration later, open the Signal and click “Edit” on any configuration section, then “Save”. Changes apply to new Waves; each Wave keeps a copy of the configuration it ran with.

1. 1Click “Run Wave” on the Signal's page (or on the Signals list, then select the Signal).
2. 2Under “Upload Audio”, choose the recording.
3. 3The Wave is queued and processed in the background. Its page updates automatically.

Long silences are cut before any processing, so a full-shift recording with long quiet stretches is fine. The spoken language is detected automatically. A Signal that has been deactivated can't run Waves until you activate it again.

A Wave moves from Queued to Running (“Structuring audio…”) and ends as Completed or Failed. Every Wave of a Signal is listed in its “Waves” section, with a status filter.

A completed Wave shows the duration (and how much of it was speech), the number of interactions (and how many were excluded) and turns.

• Audio: a player for the recording, with “Download original”.
• Conversation: the transcript by speaker and member type. Search it, show “Only extractions”, use “Play from here” on any turn, and “Export transcript”.
• Interactions: a profile of each detected conversation: duration, turns, the longest turn, each speaker's share of the talk and categories, every extracted value (click one to jump to that turn) and the conversation-level verdicts. Excluded interactions are marked.
• Results: the output table, filterable by interaction, with “Export CSV”, “Export JSON” and “Export to bucket”.

The output table always starts with these columns, followed by your extraction fields and conversation-level columns:

Recorders are organisation members who capture sessions with the Tavnit Recorder Android app and send them to a Signal, where each one becomes a Wave automatically.

1. 1Open the Signal and go to “Recorders”.
2. 2Click “Add recorder” and pick the member.
3. 3Enter a Referring ID (a free-form identifier such as an employee number, stamped on every Wave that recorder uploads) and choose the member type they record as.

A recorder can be set to Active or Disabled, edited or removed. Removing one keeps the Waves it already uploaded. Waves from the app show when they were recorded, the Referring ID and the member type.

Copy the ID from “Signal ID” on the Signal's page and post the audio as multipart file (or as JSON with file_base64 and filename):

curl -X POST https://run.tavnit.io/api/signals/SIGNAL_ID/run \
  -H "X-API-Key: $TAVNIT_API_KEY" \
  -F "file=@call.mp3"

{
  "success": true,
  "wave_id": "…",
  "signal_id": "…",
  "status": "queued",
  "audio_seconds": 754,
  "message": "Wave queued for processing."
}

Errors: 400 for a missing, empty, unsupported or too-long file, 402 when your organization can't start new work right now (contact the Tavnit team), 403/404 when the Signal isn't in your organisation. Follow the Wave in the app. See the REST API page for authentication.

See User roles.