Frequently Asked Questions

Yes. BGM Finder was specifically built to solve music recognition when human voices obscure the audio. Standard identification applications fail when speech covers the melody because voice frequencies distort the overall acoustic waveform. We pass the audio clip through an open-source neural vocal separation model (UVR5 MDX-Net) that suppresses spoken dialogue and extracts the isolated instrumental stem before running catalog fingerprint matching.

Stem previews and downloads are available exclusively for uploaded files. When you upload an audio or video file, you can isolate the background music, listen to the preview stem in your browser, and download the clean audio. For pasted video links, vocal separation runs strictly in the background to identify songs; audio previews and downloads are not provided for streamed links.

Yes, BGM Finder Online is 100% free. You do not need to create an account, enter payment information, or register an email address. There are no paid tiers or feature paywalls. The project is maintained as an independent audio utility, supported by unobtrusive display advertising.

While AI vocal separation significantly boosts match rates by stripping dialogue, certain songs cannot be identified. The most common reasons include: unreleased tracks, custom creator bootlegs, sped-up or slowed-and-reverb audio remixes, original video game background scores, and royalty-free production music from private sound libraries that are not indexed in commercial fingerprint databases like Shazam. In addition, extreme background sound effects (such as heavy gunfire, vehicle engines, or shouting) and low-bitrate compression (<64 kbps) can degrade acoustic landmarks. If a track is not identified on the first pass, try scanning an alternate timestamp or selecting an instrumental chorus.

BGM Finder supports public links from YouTube (standard videos, youtu.be URLs, and YouTube Shorts), TikTok, and Instagram Reels. Links will fail if the video is set to private, age-restricted, geoblocked, requires a user account login (such as WeChat, Douyin, or private Instagram profiles), or has been deleted. In addition, third-party platforms occasionally update anti-scraping measures or rate-limit automated server fetching. If your link fails to stream, simply record your screen or save the audio clip and drop the file directly into our upload box. Links from TikTok may fail depending on region and platform changes; uploading a screen recording or the video file is the dependable route.

All media processing is strictly ephemeral and designed with privacy at its core. When you upload a file or submit a video segment, audio is decoded directly into temporary storage (in memory when available, otherwise on disk), deleted within 30 minutes. All uploaded files, temporary slices, and isolated stems are automatically deleted within 30 minutes of search completion. No Accounts: We require no accounts and run no cross-site tracking ourselves; third-party advertising partners such as Google may use cookies as described below. We never retain copies of user media in persistent storage, we do not build tracking profiles, and we never sell or share user audio data.

Neural vocal isolation with deep neural networks is computationally intensive. Analyzing targeted 60 to 120-second windows (producing a 15-second isolated stem) ensures fast processing times while preserving shared GPU capacity for all visitors. Furthermore, long-form videos often change background music multiple times across scenes. By targeting the exact timestamp where the song is loudest and clearest, you dramatically reduce recognition ambiguity and improve fingerprint match accuracy.

Running neural AI stem demixing on dedicated GPU hardware incurs significant infrastructure and compute costs. To keep the service free for everyone without introducing paid subscriptions or registration requirements, an automated fair-use rate limit of 5 searches per hour per IP address is enforced. If you reach the limit, the interface displays the exact remaining wait time, and your search quota automatically resets as the hour expires.

You can upload audio and video files directly in the browser dropzone. Supported audio formats include MP3, WAV, M4A, AAC, and FLAC. Supported video formats include MP4 and WebM. The maximum file upload size is 25 MB. If your video file is larger than 25 MB, we recommend trimming the video to the specific 30-60 second section containing the unknown music before uploading.

Standard music recognizers listen to the combined, mixed master audio. When dialogue, commentary, sound effects, or narration occur simultaneously, the audio fingerprint is corrupted, and recognition fails. BGM Finder introduces a specialized neural pre-processing stage: before sending audio to acoustic fingerprint databases, it uses AI stem separation (UVR5 MDX-Net) to remove vocal formants and isolate the clean background accompaniment, enabling accurate identification where traditional apps fail.

Audio fingerprinting matches acoustic peaks, which can occasionally result in a false match if two tracks share identical royalty-free sample loops, drum breaks, or backing beats. If a match seems incorrect, check if the detected track samples the song you heard. You can also report incorrect matches or submit track corrections by emailing [email protected] with the video link, timestamp, and correct song details.

Yes. BGM Finder is fully responsive and functions directly in modern mobile browsers including Safari on iOS and Chrome on Android. You can paste mobile share links directly from the YouTube, TikTok, or Instagram apps, or upload video screen recordings directly from your phone's photo library.