Harmonie
Is your stream at the right level? The answer live, in decibels.
The problem
Going from one channel to the next is going from a whisper to a shout. Viewers correct with the volume knob, and end up turning down pre-emptively — at the expense of whoever was set correctly.
On the streamer’s side, nothing says so: headphones lie, the OBS meters measure peaks rather than perceived loudness, and nobody in chat says “you’re at −22 LUFS”.
What Harmonie does
You enter a channel name. Harmonie takes the public stream and analyses it against EBU R128 — the loudness standard used in radio and television.
Two things on screen. The stream graph, shown at all times: loudness over the last two minutes, with the target zone as a band — what is measured. And next to it, a verdict: a light, a sentence, the gap in decibels, and the exact move to make in OBS — what to correct.
The target is a zone, −18 to −16 LUFS integrated, with True Peak below −1 dBTP. These are not house values: they are the ones the Audio Engineering Society recommends for audio streaming — −16 for music content, −17 for mixed content, −18 for speech. Depending on what you broadcast, the whole range is right, and Harmonie says so rather than demanding a single point. The document is cited inside the application too, under “Why this target?”.
The rest waits behind a button: integrated loudness, loudness range (LRA), RMS, True Peak and points to watch. It is there for whoever wants to check, not for whoever wants to adjust.
Three questions, three measurements
“My overall level is fine” does not tell you whether your voice carries over your music. So Harmonie answers three separate questions, and you choose the one you are looking at.
- Overall level — measured continuously over the last 60 seconds. This is what viewers hear, every source mixed together. Target −18 to −16 LUFS.
- Voice — a guided 30-second measurement: you mute the other sources in OBS and read a supplied text. Same target −18 to −16 LUFS, because measured on its own your voice has to hold it by itself — it is the one viewers are listening to.
- Background — same principle, 30 seconds, sources running and you stay silent. Target −28 to −26 LUFS, that is 10 dB below the voice: present, never in front.
These guided measurements exist because Twitch broadcasts a single, already mixed track: nothing in the signal separates your microphone from your music. Muting the sources in turn is the only way to measure each one for real. The timing accounts for the stream delay; you only have to follow the instruction on screen.
In addition, the details show the gap between the speech band (300–4000 Hz) and outside the band — lows and highs, where most musical pads and game effects live. It is an immediate clue, with nothing to mute. But these are frequency bands, not separate sources: sung music counts partly within the speech band. The gap points, it does not settle — to settle, there are the guided measurements.
And above all: what to do next
A number fixes nothing. So every verdict carries its move: where to go in OBS, and what to change there. And for the full adjustment, the details lay out the ordered procedure, deduced from what Harmonie measures: which filter to add, on which source, with which values — and in which order, because that is where it is won or lost. Curb the peaks before raising the gain, put the limiter last, correct the overall level once the filters are in place.
A detail that surprises many streamers: OBS has no global volume setting. Its mixer only exposes per-source faders. The procedure takes that into account instead of sending you looking for a button that does not exist.
The suggested values come from the official OBS documentation, and every step links to it. Streamlabs being derived from it, the same filters sit in the same places there.
What matters
- Nothing to install. No plugin, no account, no access to your channel: the analysis starts from the public stream and runs on the server.
- No data collected. The channel name you ask for is not kept. No audience metrics, no forms.
- Works on someone else’s channel. Comparing your level with a channel you treat as a reference is often the fastest diagnosis.
What you should know
The measurements follow Twitch’s HLS stream: expect 5 to 15 seconds behind the live feed. That is of no consequence for setting a gain, which is judged over time rather than at an instant.
The overall-level verdict lands after about ten seconds of stream, then sharpens over a minute: a loudness average needs duration, an instantaneous value concludes nothing. Until it is refined, a banner says so — if a preroll commercial is running when you start the analysis, it skews the measurement and you have to wait for the full minute. Without a commercial, you have your answer straight away. The counter also restarts after each guided measurement: with sources muted or the microphone silent, the stream says nothing about your normal level.