Skip to main content

​
Build custom artifact detection models

Relay is a developer platform for detecting audio artifacts in Voice AI output. Train custom models on your data, then run inference to find glitches, hallucinations, unnatural pauses, and other issues in your TTS audio.

Upload Audio

Upload labeled audio files to build your training dataset

Train Models

Train custom detection models on your annotated data

Run Inference

Detect artifacts in production audio with timestamped results

API Reference

Complete REST API documentation with examples

​
Quick example

Run inference on audio to detect artifacts: 
curl -X POST https://api.relayai.dev/api/v1/inference-jobs \
  -H "X-API-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model_id": "your-model-id",
    "config": {
      "threshold": 0.5
    }
  }'
Detection results include timestamps and confidence scores: 
{
  "detections": [
    {
      "artifact_type": "glitch",
      "start_ms": 1200,
      "end_ms": 1450,
      "confidence": 0.87
    },
    {
      "artifact_type": "long_pause",
      "start_ms": 3500,
      "end_ms": 4200,
      "confidence": 0.92
    }
  ]
}

​
How it works

1

Create a Dataset

Define the artifact types you want to detect (glitch, silence, hallucination, etc.)
2

Upload Audio

Upload labeled audio files in WAV, MP3, FLAC, or other common formats
3

Add Annotations

Mark where artifacts occur in each file with start/end timestamps
4

Train a Model

Submit a training job to build a custom model from your annotated data
5

Run Inference

Upload new audio files to detect artifacts using your trained model

​
Get started

Quickstart

Get up and running in 5 minutes

Authentication

Learn how to authenticate with the API