Clean Up Meeting Recordings for Transcription

Why this matters

Meeting recordings pile up background music, notifications, and room noise. Speech-to-text models drop accuracy fast when the target signal is buried.

A clean vocal stem raises downstream ASR word accuracy measurably — often the difference between usable and unusable transcripts.

How it works

Separate vocals (with optional denoise), send the clean stem to your existing ASR provider, and compare the transcript against the raw recording.

Submit a job with an input URL (S3, GCS or HTTPS), poll the job URL, and download the rendered output. No GPU, no queues, no ffmpeg builds to babysit.

Code

import requests, time

API = "https://api.mlslabs.io/v1/vocal-separator/jobs"
headers = {"X-API-Key": "YOUR_API_KEY"}
payload = {"input": "s3://bucket/meeting.m4a", "stems": ["vocals"], "denoise": true}

resp = requests.post(API, json=payload, headers=headers)
job = resp.json()
while job["status"] not in ("succeeded", "failed"):
    time.sleep(3)
    job = requests.get(job["url"], headers=headers).json()

print("Output:", job["output_url"])

Pro tips

• Benchmark before/after on the same 3 calls to quantify the gain.
• Keep speaker diarization on your ASR side — separation does not change it.
• Batch the whole week's recordings on a schedule.

Try the meeting-cleanup flow in Postman.