Upload the beat and check tempo and key
Upload the instrumental, and Lyro reads its tempo and key in the browser in a few seconds. The vocal model also listens to the beat itself, but the readings steer the studio’s grid and the second path, so check them. Tempo is the speed in beats per minute (BPM). Key is the set of notes the music is built on, such as F minor.
- Export the beat at one constant tempo, without a guide vocal.
- Check the BPM. Detectors can return half or double the true tempo, for example 70 instead of 140. Correct it if so.
- Check the key. A detector can confuse a minor key with its relative major, such as A minor and C major, which share the same notes. A key that is a semitone out matters far more.
Bringing your own beat is on every plan, Starter included. The free BPM and key finder works on any file.
Choose a path: your lyrics or an improvised vocal
Pick the lyrics path when the words matter, and the improvised path when you want a melody idea. A topline is the main vocal melody that sits on top of a beat.
| Lyrics path | Improvised path | |
|---|---|---|
| You give | Your beat, lyrics, a voice type and a language | Your beat and a short description of the voice |
| Lyro does | A vocal model listens to your beat and sings your lyrics on it | A Lyro studio tool takes your beat as input and improvises one new vocal track that follows it |
| You get | A vocal track on its own, singing your words | An isolated vocal stem with an improvised topline |
| Model | Lyro’s vocal model; Lyro High plus separation for Turkish lyrics or when it is busy | Lyro’s studio tools |
| Credits | 7 for a beat of up to a minute, 11 for a three-minute beat | One credit per started minute, minimum 5 |
| Best for | Words that must be understood | Melodic ideas, ad-libs and textures |
Two limits. On the lyrics path the model writes its own melody for your words, and when a job takes the second path (Turkish lyrics, or the vocal model was busy for two minutes) the vocal is sung over a backing composed in your key and tempo rather than over your beat, so check the fit. On the improvised path, do not expect clear words: it improvises a melody idea, and the words are not what it is for. When the words matter, use the lyrics path, and for Turkish keep it on Lyro High, the model that sang Turkish most clearly in our blind test.
What an isolated vocal stem means
An isolated vocal stem is an audio file that contains only the voice, with no drums, bass or other instruments. A stem is one layer of a song saved as its own file. Lyro’s stem splitter separates a track into stems: every instrument on its own track in Pro mode, or vocals, drums, bass and other in Standard mode.
You get a stem, not a finished mix, so your beat stays exactly as you made it. Both parts open in the mixer lined up on bar one. A vocal from the vocal model is made on its own; one from the second path is separated from a full song, so it can carry faint traces of the instruments it was sung over. Judge it in the mix.
Mix the vocal into the beat
Four moves do most of the work: filter, compress, add space, then make room. Mixing is balancing separate tracks so they sound like one record. It uses no credits.
- High-pass filter. It removes everything below a chosen frequency. Rumble under a voice clouds the kick and bass. Start around 80 to 100 Hz, raise it until the voice thins, then back off.
- Compression. A compressor turns down the loudest moments, so quiet words are not lost. Lower the threshold until loud phrases drop by a few decibels.
- Reverb send. A send routes a copy of the vocal to a shared reverb, the effect that adds a sense of room. The dry vocal stays clear, and one control sets how much space you hear.
- Pump. Pump is tempo-synced ducking: the level dips on each beat and recovers, the breathing effect known from house music. Keep it light on a lead vocal.
Set the vocal level last: loud enough to follow every word, quiet enough that the drums keep their punch.
Master the finished track
Mastering sets the final loudness and tone of the mix. Lyro masters to a target measured in LUFS, a unit for how loud audio sounds over time. Less negative means louder.
- Pick the streaming target, -14 LUFS, for releases and social video. Pick the louder club target for DJ sets.
- The limiter holds peaks under a ceiling, so the level can come up without clipping, the harsh distortion of a signal that is too loud.
- Use A/B to switch between the master and the unmastered mix, then export WAV.
Downloads of generated audio carry a machine-readable tag that declares them AI-generated. Article 50 of the EU AI Act, on transparency for AI-generated content, applies from 2 Aug 2026 (EU AI Act, Article 50). Systems already on the market before that date have until 2 Dec 2026 to meet its marking duty (European Commission, 24 Jul 2026).
Common problems and fixes
Most problems come from a wrong tempo or key reading, so check those first.
| Problem | What you hear | Fix |
|---|---|---|
| Key clash | The vocal sounds sour against the chords | Correct the detected key. If the vocal is one or two semitones out, use pitch shift. Otherwise generate again with the right key. |
| Timing drift | The vocal starts in time and slowly slips against the beat | Look for a half or double tempo reading. For small drift, use time-stretch in the vocal tools. A beat with a changing tempo will always drift. |
| Sibilance | Harsh “s” and “sh” sounds that jump out | Turn down the top band of the vocal’s EQ a little, or try a vocal preset. An EQ raises or lowers chosen frequency ranges. |
| Unclear words | You cannot follow the lyrics | Use the lyrics path on Lyro High, shorten the lines, and generate two or three takes. |
Questions
Can AI sing my lyrics over my own beat?
Yes. On any Lyro plan, Starter included, you upload the beat, confirm its tempo and key, and add lyrics. Lyro’s vocal model listens to the beat, sings your lyrics on it and returns only the vocal, as its own track.
Will the AI vocal match the key and tempo of my beat?
Usually. The vocal model hears your beat itself; on the second path the model is given your beat’s tempo and key instead. Either can still miss, so check the fit in the mixer. Pitch shift and time-stretch handle small corrections.
How many credits does it cost?
With lyrics, 7 credits for a beat of up to a minute and 11 for a three-minute beat, whichever path makes the vocal. An improvised vocal costs one credit per started minute, minimum 5. Mixing, mastering and downloads use no credits.
Can I release the result commercially?
Every paid plan includes a commercial licence, subject to our terms and our model providers’ terms. Your own beat is your own work. We do not promise copyright protection for generated parts. See Who owns AI-generated music?