While the call is live: who spoke, what they said, without the personal data.
Supervisors cannot wait for the recording to land. They need the transcript as the conversation happens, with speakers marked and account numbers already removed.
What breaks in the contact center
The transcript arrives after the call
Coaching that waits for the file is coaching on yesterday. The supervisor needs the words while the caller is still on the line.
Speaker turns are a second pipeline
Who spoke is often a separate system. Two integrations, two failure modes, and the live view falls apart.
Personal data in the live stream
Card numbers and national IDs in a raw transcript are an incident. Redaction has to happen in the same stream, not in a later batch.
What the API returns
Partial transcripts as the caller talks
Streaming speech-to-text with partial results. The supervisor sees the sentence forming, not a file after hang-up.
Speaker IDs on the same stream
Turns marked SPEAKER_00, SPEAKER_01 on the words themselves. No second vendor for diarization.
Configured entities replaced in place
Card numbers, national IDs and addresses become tokens in the live text. The compliance export is the same stream.
What a live turn looks like
This is a transcript with speakers and a redacted span — a visible result, not a category label. Copy the call, then run it with a key.
- SPEAKER_00
How can I help today?
- SPEAKER_01
I'd like to update the card ending [REDACTED].
{
"text": "I'd like to update the card ending [REDACTED].",
"speakers": [
{ "id": "SPEAKER_00", "role": "agent", "text": "How can I help today?" },
{ "id": "SPEAKER_01", "role": "caller", "text": "I'd like to update the card ending [REDACTED]." }
]
}Sample response. The next step is to run this call with your own audio.
curl -X POST https://api.brainiall.com/v1/stt/transcribe/base64 \
-H "Authorization: Bearer $BRAINIALL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"audio":"<base64-wav>"}'What changes for the operator
Flags in the moment
A supervisor prompt can fire while the caller is still talking. That is a different job from a nightly quality batch.
The compliance copy is already clean
The stream that coaches the floor is the stream that leaves the building. No second pass to strip personal data.
What we publish, and what we do not
We do not invent seat counts or unpublished contact-center outcomes. Speech quality numbers live on the audio product page. Open the published audio numbers
Transcribe a call of your own
Create a key and run the speech-to-text call. The next screen is this job: a transcript, not a text label.
The next screen is the same job: a key, then this call.