Diffio API

Build recorded-speech restoration into your product.

Add cloud batch restoration for recorded speech with an API key and the Diffio SDK. Review the API pricing page, then start with the developer quickstart.

Cloud batch workflow

From an uploaded recording to restored speech audio

Use the explicit resource workflow when your backend needs control over each stage. The developer quickstart includes complete Python and Node examples.

  1. Create and upload a project

    Create a project for the recorded-speech file. The SDK uploads the file during project creation.

    Create project endpoint

  2. Start a generation

    Create a generation with diffio-4.5-flash or diffio-4.5-pro.

    Diffio 4.5 Flash generation endpoint

  3. Wait for completion

    Poll generation progress, or configure webhooks to receive completion and failure events.

    Progress endpoint Webhooks guide

  4. Download restored speech audio

    Request the completed generation download and save the returned speech audio.

    Download endpoint

Diffio processes uploaded recordings as cloud batch jobs rather than as a real-time streaming API. For supported video inputs, API processing restores the speech soundtrack while the picture and other visuals remain unchanged.

Python SDK helper

For a compact backend integration, restore_audio performs the same cloud batch workflow in one SDK helper call.

Python SDK, one SDK helper call

Restore a recorded-speech file with restore_audio

Load the API key from the environment and write the returned MP3 bytes to disk.

Explore SDK docs
restore_amelia.py
from dotenv import load_dotenv
from diffio import DiffioClient
import os

# pip install diffio python-dotenv

load_dotenv()

client = DiffioClient(apiKey=os.environ["DIFFIO_API_KEY"])

audio_bytes, info = client.restore_audio(
    filePath="amelia_earhart.wav",
    model="diffio-4.5-flash",
    onProgress=lambda progress: print(progress.status),
)

if info["error"] or audio_bytes is None:
    raise SystemExit(info["error"] or "Restore failed")

with open("amelia_restored.mp3", "wb") as handle:
    handle.write(audio_bytes)

Before and after demo

Listen to an Amelia Earhart speech-restoration comparison

Switch between the source broadcast and a completed Diffio restoration while it plays. Use before-and-after listening on representative recordings to evaluate artifacts and intelligibility for your own integration; results depend on the source.

This is a saved Diffio output, so the model and settings used for it aren't listed. Try the API on your own recordings to hear current results.

Loading waveform...
Elapsed time 0:00 of 0:31
Loading your audio…Playback controls turn on when it's ready.

Available recorded-speech models

Review the current model cards for selection details and API pricing for current rates.

Diffio 4.5 Flash

Multilingual recorded speech

Model IDdiffio-4.5-flash

Diffio 4.5 Pro

Multilingual recorded speech

Model IDdiffio-4.5-pro