Speaker diarization API from $0.02 an hour

Speaker labels on their own

Already have a transcript? Speaker labels alone cost $0.02 an hour of audio. Each speaker turn has its speaker, start and end, as JSON or RTTM.

Or a transcript with speakers

A speaker transcript costs from $0.07 an hour. Every segment has its speaker and timestamps, as JSON, SRT or VTT.

Avencera serves OpenAI’s transcription API, so the official OpenAI SDKs work. Set speaker_labels to true on the request.

Whole recordings, fast

OpenAI SDK requests take files up to 2 hours and 100 MB. Longer recordings, up to 24 hours, go through a job, not cut into short clips first.

speakrs runs its own GPU kernels written in Rust. On the same GPU it is about 60× faster than pyannote in Python, at 7.0% diarization error rate against pyannote’s 7.2% on the VoxConverse dev set.

A speaker transcript in Python

from openai import OpenAI

client = OpenAI(base_url="https://api.avencera.com/v1", api_key="YOUR_AVENCERA_KEY")

with open("meeting.mp3", "rb") as audio:
    transcript = client.audio.transcriptions.create(
        model="avencera/parakeet-tdt-0.6b-v2",
        file=audio,
        extra_body={"speaker_labels": True},
    )

Questions

What is speaker diarization?

Speaker diarization splits a recording by who is speaking. The result is a list of turns, each with a speaker label and its start and end time.

Which diarization model does Avencera run?

pyannote’s open community-1 pipeline, run through speakrs, an open-source Rust implementation that matches pyannote’s published error rates.

How much does the speaker diarization API cost?

Speaker labels alone cost $0.02 an hour of audio, and a transcript with speakers costs from $0.07 an hour. You buy prepaid credit, with no subscription and no minimum.