Call Metadata

Call Metadata

Uspeech analyzes what happens inside a call. Everything around the call — when it happened, which agent handled it, whether it was inbound or outbound, which campaign it belonged to — comes from your own systems. There are three ways to provide it:

• A CSV manifest for a whole project — the fastest way to cover a batch of recordings
• The Edit call metadata dialog on a single file — for fixing or filling in one call
• The API, either at upload time or afterwards — see Call Metadata API

All three write to the same place, so you can mix them freely.

Why it matters

Without metadata, Uspeech falls back to the file’s upload time and to the agent name the AI hears in the conversation. That still produces a report, but:

• The weekly trend and the date filters use the upload date instead of the call date
• The shift comparison cannot classify calls (only calls with real date/time are counted)
• Agent performance relies on the name spoken in the call, which varies between recordings
• Direction and campaign are unavailable as filters

Providing metadata fixes all of these at once.

---

Uploading a manifest (many calls)

1. Click Conversations on the navigation bar
2. Select your call center project
3. On the Files tab, click Call Metadata (CSV)
4. Choose your CSV file and click Upload

The dialog reports how many files were matched, how many rows are still pending (no matching file uploaded yet), and any rows that could not be read.

---

CSV format

One row per call, with a header row:

Column	Required	Description
filename	Yes	Name of the audio or transcript file as uploaded
call_datetime	Yes	When the call took place
agent_id	No	The agent’s identifier in your own systems
agent_name	No	The agent’s display name
direction	No	inbound or outbound
caller_number	No	Phone number of the customer
campaign	No	Campaign or queue name

Example:

filename,call_datetime,agent_id,agent_name,direction,caller_number,campaign
call_001.mp3,2026-06-15 09:12:00,A-06,Ana Diaz,inbound,+34600111222,Retention
call_002.mp3,2026-06-15 09:41:00,A-10,Luis Gomez,outbound,+34600333444,Winback
call_003.mp3,2026-06-15 10:02:00,A-06,Ana Diaz,inbound,+34600555666,Retention

Accepted date/time formats:

• 2026-06-15T09:12:00 (ISO 8601, with or without a time zone)
• 2026-06-15 09:12:00 or 2026-06-15 09:12
• 15/06/2026 09:12:00 or 15/06/2026 09:12

Values without a time zone are interpreted in the server’s time zone.

---

How rows are matched to files

1. Exact filename match first (call_001.mp3 → call_001.mp3)
2. Name without the extension, case-insensitive, as a fallback — so a manifest that lists call_001.wav still matches an uploaded call_001.mp3

Two useful consequences:

• Order doesn’t matter. Rows that don’t match anything yet are kept on the project and applied automatically to files uploaded later. You can upload the manifest before, during, or after the recordings.
• Transcripts count as calls. SRT and VTT uploads are matched exactly like audio files.

You can upload a manifest more than once; later uploads add to and update the stored rows.

---

Rows that are skipped

A row is rejected — and reported back in the dialog — when:

• filename is empty
• call_datetime is missing or not in one of the accepted formats

Everything else is optional: unknown or empty optional columns are simply ignored. A direction value that isn’t inbound or outbound is dropped, and the rest of the row still applies.

---

Editing one call

In a call center project the files table shows two extra columns — Call time and Agent — so you can see at a glance which calls already carry metadata. A call time shown in grey is the file’s upload time, used as a placeholder because no real call time has been provided yet.

To change a single call, open the ⋯ menu on its row and choose Edit call metadata.

The dialog is available on audio files and on SRT/VTT transcripts. A few things worth knowing:

• Empty clears. Saving a field left blank removes the value that was there. This is the opposite of the CSV manifest, where a blank cell simply leaves the existing value alone.
• Call time can’t be emptied. Clear it and the call falls back to the file’s upload time.
• A later manifest wins. If you upload a CSV manifest afterwards, its rows overwrite what you typed here. Edit individual calls after the bulk load, not before.
• Editing metadata never re-runs analysis. Scores, topics and sentiment are left exactly as they are; only the descriptive fields change.

Tip

Use the dialog for corrections and one-off gaps. If more than a handful of calls are wrong, fixing the export from your own system and re-uploading the manifest is faster.

---

Agent identity

Uspeech keeps one agent record per person and links every call to it:

• If you provide agent_id, it is the key — renaming the agent later keeps the history intact
• If you provide only agent_name, agents are matched by name (case and spacing are normalized)
• If you provide neither, the agent name the AI extracted from the conversation is used as a fallback

Metadata always wins. A name extracted from the audio never overwrites an agent supplied by the manifest or the API.

Tip

Use stable agent_id values from your ACD or workforce management system. It is the only way to keep agent statistics correct across name spellings, married names, and duplicate names.

---

Sending metadata through the API

If you upload recordings programmatically, you can attach the same fields to each file at upload time instead of maintaining a CSV, change them afterwards, and read back what a call currently carries — see Call Metadata API.