Find a song by its sound

Hear a song you don't know? Record a few seconds of it, or upload a voice note or audio file. The song finder listens to several parts of the clip, only answers when they agree, and then sends you the full song with its lyrics.

Try it here

On the web you can record up to 15 seconds or upload a file up to 50 MB. In Telegram you can also send voice notes, videos and links, and get the full song downloaded straight into the chat.

Open Song Finder by Sound in Telegram Free · no sign-up · works on phone and desktop

How to use it on this website

  1. Press Record and hold your phone or computer near the music for about 10–15 seconds – or press Upload and choose an audio or video file.
  2. Allow microphone access if your browser asks. The recording stays on your device until you press send.
  3. Wait a few seconds while the clip is analysed.
  4. See the song's title, artist and cover, with buttons to download it, read its lyrics or open it in the bot.

How to use it in the Telegram bot

  1. Open @MoeinimyDL_bot and hold the microphone button to record a voice note of the song playing – or forward any audio file.
  2. Under your voice note or file, tap Identify the music.
  3. The bot listens to several windows of the clip and announces the song only when two of them agree. If it is unsure, it shows the candidates it heard for you to pick.
  4. The full song arrives with its cover art, tags and a Lyrics button.

Open Song Finder by Sound in Telegram →

How the song finder listens

A single guess from a single moment of audio is unreliable: talking, an effect or a quiet intro can make any recognition engine answer with something unrelated. So instead of asking once, the song finder cuts your clip into several short windows across its length and prepares each one in more than one way – as recorded, with the level evened out, and with the bass and hiss filtered away. It then asks about them in parallel.

A song is announced only when two independent windows agree on it. If only one window hears something, or two hear different songs, you get the candidates as buttons and pick – rather than being told a confident wrong answer.

Silent stretches are skipped instead of wasting a lookup, and the whole search is time-boxed so you are not left waiting on one slow reply.

More than one engine

Shazam is fast and has an enormous catalogue, but no catalogue has everything – and Shazam's gaps are largest for Persian and other regional music. In Telegram, when the first pass has not settled it, the bot races other engines alongside: ACRCloud, AudD and AcoustID, which fingerprint the recording differently, and a lyrics engine that transcribes the words being sung and searches for them. Their answers are merged by title, and agreement between engines counts just like agreement between windows.

What happens next

Knowing the name is half of it. In Telegram the bot goes on to find that song – the studio recording, checked against the catalogue – and sends it to you as a tagged file with its cover art and a Lyrics button. On the web you get the same answer with buttons to download it, read the lyrics or open it in the bot.

Where the music is in a video

If the song is playing in a video – a reel, a TikTok, a YouTube clip – you don't need to record it. See find a song from a video: send or upload the video itself and the soundtrack is identified directly.

Tips for a good match

  • Get close to the speaker and away from conversation.
  • Record the chorus if you can – it is the most distinctive part.
  • 10–15 seconds is enough; more does not hurt.
  • For live performances and covers, recognition finds the original only when the performance is close to it.

Formats, sizes and limits

On the web In the bot
Input Microphone (up to 15 s) or a file up to 50 MB Voice note, audio, video, video note or a link
Longest clip analysed The first minute, sampled The whole file, sampled across its length
Engines Shazam, plus free engines Shazam, racing ACRCloud, AudD, AcoustID and a lyrics engine where they are enabled
What you get Title, artist, cover, lyrics, download The full song file, tagged, with lyrics
Confidence Two agreeing samples, or a list to pick from Two agreeing samples, or a list to pick from

Frequently asked questions

How do I find a song by its sound?

Record 10–15 seconds of it playing, or upload a clip. The song finder fingerprints the audio, compares it with millions of recordings and tells you the title and artist – and in the bot, sends you the whole song.

Is this the same as Shazam?

Shazam is one of the engines used. The bot can also ask ACRCloud, AudD and AcoustID at the same time and combine their answers, which catches songs Shazam alone misses – Persian music in particular.

Can it recognise humming or singing?

No. Recognition matches recordings, so it needs the song itself playing – from a speaker, a video, the radio. Humming and singing do not match the original fingerprint.

How long should the recording be?

10–15 seconds of clear music is plenty. Longer is fine – several parts are sampled – and very short clips (under 5 seconds) or clips where people talk over the music are the hardest.

Why did it not find my song?

Usually the music is too quiet under speech or noise, the clip is very short, or the song is not in the recognition catalogues (common for very new or very local releases). Try a longer, cleaner part, or type what you know into the song downloader.

Why does it sometimes give me a list instead of an answer?

A wrong answer is worse than none. The song finder only announces a song when two separate parts of the clip agree. When they don't, you see the candidates it heard and choose.

Is my recording stored?

No. The clip is used to identify the song and deleted straight away. See the privacy page.

See every tool →