Skip to main content
Speech Recovery improves the intelligibility of speech in degraded or difficult recordings while staying faithful to the original audio. Both models enhance the existing signal rather than regenerating it, so the output avoids the synthetic artifacts common to generative “enhance speech” tools. They’re designed to work on low-resolution and heavily degraded audio, making it possible to recover understandable speech from recordings that would otherwise be difficult to use. The suite offers two operations:
  • Denoise (speech_denoise) — removes background noise, hum, and interference while preserving the recording’s natural acoustics. 1.5 credits / min.
  • De-reverb (speech_dereverb) — removes reverberant room signal (reflections, echo) to improve intelligibility in echoey spaces and ease downstream post-production. 2 credits / min.
Speech Recovery fixes bad recordings; it’s a separate family from Dialogue, which extracts the dialogue stem from an already-clean film/TV mix. Reach for Speech Recovery when speech is present but hard to understand due to noise, limited bandwidth, recording artifacts, or room reverberation.

​
Which model to use

  • speech_denoise — for ultra-noisy, low-quality, low-resolution recordings, or recordings with hum, where you need the words clear. Especially helpful for improving “background” speaker intelligibility. Preserves the natural ambience of the recording.
  • speech_dereverb — when you want to cleanly isolate and denoise the speech but also remove echoey rooms and reverberant spaces, producing a drier, closer-sounding result. Removes room character by design, so it’s opt-in.

​
Create a Task

import requests

API_KEY = "your_api_key"
HEADERS = {"Content-Type": "application/json", "x-api-key": API_KEY}

response = requests.post(
    "https://api.audioshake.ai/tasks",
    headers=HEADERS,
    json={
        "assetId": "your_asset_id",
        "targets": [
            {"model": "speech_denoise", "formats": ["wav"]}
        ]
    }
)

task_id = response.json()["id"]
print(f"Task created: {task_id}")
To remove reverberant room signal instead, use speech_dereverb as the model. Check Task status to monitor progress and download results, or use webhooks to be notified when each target completes.

​
Use cases

  • Broadcast / journalism — field recordings in noisy environments, sports broadcasts with heavy crowd noise, and remote interviews recorded outside studio conditions (including echoey rooms).
  • Film / television — location audio affected by weather or environmental noise, reverberant interiors, and unscripted or documentary material where ADR is undesirable. De-reverb produces drier dialogue for ADR matching and editing.
  • Public safety / forensics — emergency call recordings, low-quality surveillance or body-cam audio, reverberant recordings from rooms or vehicles, and audio evidence review.
  • Healthcare / legal transcription — improving clarity of recorded consultations and making speech easier to transcribe in noisy or echoey environments.

Multi-Speaker Separation

Separate individual speakers before recovery.

Dialogue Separation

Isolate dialogue from a finished film/TV mix instead.