Live translation and captions — run from Australia, Europe and the United States
SpanVoice Try it free

Setup guide

Set a clean audio level for live translation.

A clear speech feed gives live translation a better starting point. This guide shows where to adjust the level, how to check loud and quiet speakers, and how to test the recognised words before the event.

A hand adjusts the output-level knob on a small mixer while a microphone feeds its first channel and a blue USB cable connects it to a laptop showing a green level meter.

Where the level is set

The level is set before the sound reaches SpanVoice, and the meter in the desk app shows the result. Set it with the loudest speaker, correct any clipping at or before the stage where it happens, then check that the feed is quiet when nobody is speaking.

Speaker loudest expected
Translation send mic channels clean first
Interface its input level
Desk app meter and transcript
Not every room has every step: with no desk, the first control may be Windows sound settings.

Set a clean audio level in six steps

Do it once with the room empty, then confirm with a real sentence.

  1. Find the earliest place you can change the level

    The desk app has no level control of its own. It shows you what Windows gives it, so the level is changed upstream, at the stage that needs it:

    1. On the sound desk, if the feed comes from one. First check that each speech microphone channel is clean, with its own gain not clipping. Then set the dedicated translation send, the one only SpanVoice listens to, to a suitable level. Adjust that feed, not the main room volume.
    2. On the audio interface, if there is one: its input gain, or whatever level control that model has.
    3. In Windows sound settings, for the recording level of the device. This is the place for the computer’s own microphone, or a USB microphone with no knob.

    If something clips, correct the level at or before the stage where it clips. Turning a distorted signal down further along does not repair it: lowering a send does not help if the microphone preamp is already clipping, and lowering the interface gain is the right fix if that is where it clips. A mixer’s USB feed may be taken before its output control, so check its manual before relying on that knob to change the USB level.

    Speaker the loudest you expect
    Translation send mic channels clean first
    Interface gain, if it has one
    Desk app the meter shows the result
    Correct the level at or before the stage that clips; the meter in the app only reports what arrives.
  2. Set it with the loudest speaker, in their real position

    Open the desk app and choose the input. The meter runs as soon as a device is chosen, before you start streaming, so nothing is sent while you do this. Have the person who will speak loudest stand where they will actually speak, with the microphone they will actually use, and talk the way they will on the day.

    Adjust until the meter moves clearly into the green on every sentence and the two red segments at the top stay dark, with room to spare for a louder moment. That room is headroom: the space between normal speech peaks and the level where the signal overloads. It does not need to be loud; steady and clean matters more than high.

    The meter shows the average level of the sound arriving from the input, measured in short slices, on a scale from about -60 dB at the bottom to 0 dB, the loudest the input can carry, at the top. Each segment is roughly 4 dB. The two red segments light only when that average is within about 6 dB of the top, which for speech usually means its loudest moments have already hit the ceiling. Treat red as a sign to check for clipping. Check the desk and the interface for clipping as well: a healthy-looking meter in the app does not prove the signal was clean earlier in the chain.

    A quiet speaker who arrives later will sit lower, and that is fine. A loud one who pushes the meter into the red is the one you are guarding against.

  3. Stop talking and check what the meter does

    Ask them to stop. The meter should fall almost to nothing. If it keeps moving or sits part way up, something other than the speaker is in the feed. Work out what by taking things away one at a time:

    • Other microphones that are open but unused. Mute them on the desk.
    • Music, playback or effects that have reached the send. Take them out of it.
    • Room noise such as air conditioning, a projector fan or an audience before the start. Bring the microphone closer to the speaker rather than raising the level.
    • Hum or buzz from the cable path. See the mixer guide for the cable and ground checks.
    • The input’s own noise, when the gain is turned a long way up to make up for a quiet source. Raise the level at the source instead.

    Mute the translation send for a moment. If that mute controls the output SpanVoice is taking, a drop in the meter helps locate the noise upstream, in the mix; if it does not drop, check the interface and the Windows input.

  4. Cover the other voices in the room

    The loudest speaker sets the ceiling. Then check the rest, because one setting has to work for all of them:

    • Someone who speaks softly or turns away from the microphone. If they are too low, ask them to speak closer or use a microphone that stays near them; raising the level for them can push the next loud speaker into the red.
    • A handheld microphone that moves between the mouth and the chest. Ask the speaker to hold it steadily.
    • Guest speakers and new microphones. Each has to be in the feed, at a similar level to the others. A microphone that was added to the room mix but not to the SpanVoice send is never heard.

    Where one microphone is much louder or quieter than the rest, balance it on the desk rather than changing the whole feed.

  5. Check the words, not only the meter

    A good meter is necessary, not proof. Press Start streaming, say a sentence with a name and a number in it, and read the words on your own phone from the audience link. In the desk app, open Transcript: the Heard in the room column shows what is being recognised. If words are missing or garbled while the meter looks right, check the speaker’s position and the microphone, and check the desk and interface for clipping, before changing the level again.

    With nobody to speak, switch the Input panel to Test file and play a recording. That checks the app and the streaming path, not the desk, so it tells you nothing about the level on the day; do the live check as well.

    If the speech is clear but a name is repeatedly recognised incorrectly, check the spoken-language setting and add the name to the Glossary, then test again.

  6. Leave it alone, and keep an eye on it

    Once it is right, leave it and save the working setup as your starting point, including the desk scene with the translation send in it. A recalled scene or a new microphone is the usual reason it changes. Before each event, check the actual microphones and speakers, and confirm that the feed remains clean.

    During the event, glance at the meter now and then. The meter keeps running when you press Pause audio, so it still shows what is arriving. If someone is speaking and the meter stays still, check the selected input and the audio path. If it has climbed into the red since you set it, someone may have raised a level upstream. Look before you change anything in the app.

What the meter is telling you

Read it with someone talking, then with nobody talking.

What you seeWhat it usually meansWhat to do
Nothing lit, reading shows a dashIf someone is speaking, the sound may not be reaching the computer.Check the selected input and the audio path: cable, send and chosen device. The help centre page Nothing is being heard goes through it in order.
A segment or two flickering while someone talksSound is arriving at a low level.Check in the transcript whether the words are understood before adding gain, and move the microphone closer.
Clearly in the green on every sentence, falling back between wordsUsually right, if the red stays dark and there is room to spare.Check the words in the transcript, then leave it.
Reaching the redThe signal needs checking: the loudest moments are probably at the ceiling.Lower the level at or before the stage that clips, and check the desk and interface for clipping.
Part way up with nobody talkingThe feed is carrying something other than speech, or noise.Take away open microphones, music and effects, and check for room noise and hum.

Troubleshooting the level

Match the symptom to the likely cause, then change one thing and listen again.

Words are missing from sentences

The level may be too low, so quiet words are lost, or the speaker is too far from the microphone. Move the microphone closer before raising the level.

Words come out garbled

Check first for red on the meter and for clipping on the desk and the interface. Lower the level at or before the stage that clips, not afterwards.

Noise turns into words

The feed carries room noise, music or an open microphone. Take them out of the send, mute microphones nobody is using, and use Pause audio during music.

One speaker is much quieter than the others

Balance that microphone on the desk, or ask the speaker to come closer. Raising the whole feed for one voice risks the next loud one.

The level changes between events

A scene was recalled without the send in it, a microphone was added or a level was moved. Save the scene with the send in it and check the meter before people arrive.

The meter looks right but a name is wrong

If the speech is clear but a name is repeatedly recognised incorrectly, check the spoken-language setting and add the name to the Glossary, then test again.

Still unsure

Questions about the level

Describe what the meter does and we will tell you where to look.

Talk to us

Is a louder level better?

No. A feed that is overloaded loses words, and a feed that is steady and clean with room to spare gives the best starting point. Louder only helps when the sound was too quiet to start with.

Does turning the level up make the words appear faster?

No. Turning up the gain is not a way to speed up the words. Aim for clear speech without clipping. How quickly words appear also depends on the speech, the processing and the internet connection.

Does the level change what it costs?

Input gain is not a pricing setting. What you pay is set out on the pricing page.

Can I change the level in the desk app?

No. The app shows what Windows delivers. Change it on the desk, on the interface, or in Windows sound settings for a device with no other control.

Which control do I use if I have a desk, an interface and Windows?

Check the microphone channels are clean first, set the translation send to a suitable level, then adjust the interface input gain. Use the Windows level to finish. If something clips, correct the level at or before that stage; turning it down later does not repair it.

Do I need to set it before every event?

A complete reset is usually unnecessary if the feed comes from a saved translation send. A brief soundcheck is still needed: check the actual microphones and speakers, and that the feed remains clean, before people arrive.

What if I only have the computer’s own microphone?

Set the recording level for it in Windows sound settings, and check Windows is allowing desktop apps to use the microphone. Keep it close to the speaker; a laptop microphone across a room picks up the room.

Set the level and read the words on your own phone.

The free credit lets you test the setup at your own desk. No credit card needed.

A$10 of free credit when you sign up, no credit card needed. Credit never expires — use it this week or next year.