Skip to main content

Documentation Index

Fetch the complete documentation index at: https://runinfra.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

A transcription pipeline takes audio files of any length and returns a transcript with timestamps, speaker labels, and optional PII redaction. RunInfra ships the recipe with Whisper (large-v3, distil-large-v3, or turbo) for ASR, pyannote-style diarization for speaker labels, and a small classifier for PII redaction.

​
Architecture

Audio file (mp3 / mp4 / wav / m4a / webm)
  -> Whisper ASR (large-v3 or distil-large-v3, FP8)
  -> Diarization (speaker turns, optional)
  -> PII redaction pass (names, emails, phone numbers, optional)
  -> Transcript with timestamps + speaker labels + redaction markers
Long-form audio is chunked with overlap, transcribed in parallel batches on the GPU, then stitched with timestamp alignment. The whole stack runs on one L40S for files under 90 minutes.

​
What you get out of the box

  • OpenAI-compatible /v1/audio/transcriptions endpoint (multipart upload)
  • response_format: json, text, srt, vtt, verbose_json
  • Speaker labels via diarization (set diarize=true)
  • PII redaction with replacement tokens (set redact=true)
  • Long-form support: files up to several hours, chunked and stitched

​
Example prompt

In Pipes: 
Build a transcription pipeline for our recorded support calls.
Use Whisper large-v3 with diarization and PII redaction.
Output should be SRT subtitles plus a JSON transcript with speaker labels.

​
Quick example

from openai import OpenAI

client = OpenAI(base_url="https://api.runinfra.ai/v1", api_key="YOUR_RUNINFRA_API_KEY")

with open("call.mp3", "rb") as f:
    transcript = client.audio.transcriptions.create(
        model="your-pipeline-id",
        file=f,
        response_format="verbose_json",
        extra_body={"diarize": True, "redact": True},
    )

for segment in transcript.segments:
    print(f"[{segment['speaker']}] {segment['text']}")

​
Output shape

{
  "text": "Hello, I'm calling about [REDACTED_EMAIL] order...",
  "segments": [
    { "start": 0.0, "end": 2.1, "speaker": "Speaker 1", "text": "Hello, I'm calling about [REDACTED_EMAIL] order..." }
  ],
  "language": "en"
}

​
Deeper details

See runinfra.ai/use-cases/transcription for the marketing page with per-minute cost math and supported audio formats.