Skip to main content
Contact Us

Media & Entertainment

AI Audio Recognition

Upload a file, record your voice, or pick a sample clip and find out whether it is speech or a sound effect. Two AI models analyze each clip at once, and the more confident result wins.

Audio Recognition Demo with sample clips, a player, and the recognized Crow sound effectTry the demo

Upload an audio file up to 1 MB or record up to 2 minutes. The demo runs on a test server, so analysis may be slower than in production.

How it's builtPython, FastAPI, PyTorch, OpenAI Whisper, Next.js, React, TypeScript

A Next.js and React interface sends audio to a FastAPI service. Two models run in parallel: OpenAI Whisper transcribes speech and gives a confidence score, and an AudioSet classifier names the most likely sounds. A confidence rule then picks the final result: speech with its transcript, or a sound effect with its label.

Features

The demo input panel with microphone recording, audio file upload, and built-in sample clips

Pick, upload, or record.

Start from sample clips, upload an audio file up to 1 MB, or record up to two minutes with your microphone.

A real speech sample selected in the audio demo, with its player and recognized transcript

Two models, one answer.

Whisper listens for speech while a sound classifier listens for everything else. Both run at once, so the wait depends only on the slower one.

The audio player and recognition result identifying the built-in sample as a Crow sound effect

See the result at a glance.

The player labels each clip as speech or a sound effect and shows the transcript or the top sound label.