Documentation

Noise Suppressor

Upload a file, record from your microphone, or load the built-in noisy sample clip. Pick a model, set the noise reduction limit and output format, then press Remove noise. The original and the cleaned audio play side by side, and the cleaned file downloads in one click.

API explorer & keys

The API explorer lists every service published to the gateway, with a request runner and code snippets. Services registered in the gateway appear there automatically, without a studio release.

Base URL & auth

Every API is reached through the gateway at https://internal.audiclabs.com/v1/{service}/{endpoint}. Send your access token in the Authorization header as a bearer token, or as X-API-Key. Tokens are scoped: a token granted noise-suppressor:* can call every Noise Suppressor endpoint and nothing else.

Authorization: Bearer mw_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
X-API-Key: mw_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

POST /v1/noise-suppressor/denoise

Upload one audio file as multipart/form-data and receive the cleaned audio.

curl -X POST https://internal.audiclabs.com/v1/noise-suppressor/denoise \
  -H "Authorization: Bearer $KEY" \
  -F file=@noisy.wav \
  -F model=audic-2 \
  -o clean.wav

POST /v1/noise-suppressor/denoise/batch

Clean up to 25 files in one call with the same settings. Repeat the files field once per file; the other fields are the same as /denoise. The response is denoised.zip with every cleaned file plus manifest.json. A file that fails is listed in the manifest with its error and does not stop the rest of the batch.

curl -X POST https://internal.audiclabs.com/v1/noise-suppressor/denoise/batch \
  -H "Authorization: Bearer $KEY" \
  -F files=@call-1.wav \
  -F files=@call-2.mp3 \
  -F files=@interview.m4a \
  -F model=audic-2 \
  -o denoised.zip

import requests

files = [("files", open(p, "rb")) for p in ["call-1.wav", "call-2.mp3", "interview.m4a"]]
r = requests.post("https://internal.audiclabs.com/v1/noise-suppressor/denoise/batch",
                  headers={"Authorization": "Bearer YOUR_TOKEN"},
                  files=files, data={"model": "audic-2"}, timeout=900)
r.raise_for_status()
open("denoised.zip", "wb").write(r.content)

{
  "model": "audic-2", "output_format": "wav",
  "files": 3, "succeeded": 3, "failed": 0, "processing_s": 7.9,
  "items": [
    { "index": 0, "input": "call-1.wav", "status": "ok",
      "output": "call-1_denoised.wav", "bytes": 312364,
      "stats": { "duration_s": 9.77, "processing_s": 2.61, "rtf": 0.27 } }
  ]
}

GET /v1/noise-suppressor/models

Lists every model with its sample rate, description and whether it is installed.

{
  "models": [
    { "id": "audic-2", "label": "Audic 2 · Balanced", "sample_rate": 16000,
      "description": "Balanced quality and speed. Recommended default.",
      "available": true, "size_mb": 10.2 }
  ],
  "default": "audic-2"
}

GET /v1/noise-suppressor/sample

Returns a short noisy 16 kHz speech clip for testing.

GET /v1/noise-suppressor/health

Service status, version and installed models. Free to call; use it for monitoring.

Models

Response headers

Errors & limits

Gateway errors share one JSON shape, so clients can branch on error.type:

{"error": {"type": "scope_denied",
           "message": "This token is not scoped for 'noise-suppressor:denoise'.",
           "request_id": "req_01J7Z…"}}

Support

Use the Feedback button in the studio header to report a bug or suggest an idea. For access tokens and workspace changes, contact your workspace admin. Live component health is on the status page.