Translator — vMix Master

Translator lowers the program audio while a translator, or interpreter, is speaking, and brings it back up when they stop. It is a sidechain: it watches the level of the translator's microphone and adjusts the volume of another bus, the chain bus, that carries the program audio.

It is built for a stream where an interpreter speaks over the original audio. Without it, the operator has to ride a fader all event.

Set up the audio

The script works with this routing. The numbers are the defaults.

The program audio goes to the chain bus, B by default. Set CHAIN_BUS to the bus you use.
The translator's input goes to bus C only, with Master switched off for that input.
Add a sound delay of 800 ms to bus C. The audio is mixed 800 ms late, which gives the script time to lower the program before the translation is heard.
Send both buses to the master.

Translator input ─┬─► level meter ─► Translator script
                  └─► Bus C (800 ms delay) ─┐
                                            ├─► Master
Program audio ─► chain bus (B), set by script ┘

The translator's input is the first of these that exists:

An input with Translator in its title.
The first input of type Audio.
The first input of type VideoCall.

Add Translator to the title of the input to be sure the right one is used.

Set up the script

Add the script to the vMix that mixes the audio, see adding a script.
Check CHAIN_BUS, and the times below if your setup differs.
Start the script.

If it cannot find an input, it prints No mic / vMix Call input detected. and looks again after five seconds.

Settings

All times are in milliseconds.

Setting	Default	Meaning
`LOOP_TIME`	`50`	How often the script checks the translator's level
`VOICE_THRESHOLD`	`0.1`	The level that counts as speech. `0.1` is about -20 dB. See Audio levels
`CHAIN_BUS`	`"B"`	The bus whose volume is adjusted, `"A"` to `"G"`
`VOLUME_DOWN_TIME`	`900`	How long the fade down takes when the translator starts
`VOLUME_REDUCED`	`50`	The volume of the chain bus while the translator speaks, 0 to 100
`SILENCE_LIMIT`	`2000`	How long the translator must be silent before the volume comes up the first time
`VOLUME_UP_TIME`	`1000`	How long the first raise takes
`VOLUME_FULL`	`75`	The volume after the first raise
`SILENCE_LIMIT2`	`4000`	How long the translator must be silent before the volume comes up the second time
`VOLUME_UP_TIME2`	`2000`	How long the second raise takes
`VOLUME_FULL2`	`85`	The volume after the second raise

VOLUME_DOWN_TIME = 900 fits the 800 ms delay. If the translator goes straight to the master without a delay, use 350 so the program is already low when the voice is heard.

How it works

Chain bus volume
| ──────●                                   ●─────
|        \                                 /
|         \                               /
|          \                             /
|           \                 ●──────●
|            \                /
|             ●────────●──────●
+----+----+----+----+----+----+----+----+----+----→ Time
          T1          T2     T3

T1. The translator starts speaking, meaning the level is above VOICE_THRESHOLD and the input is not muted. The chain bus fades down to VOLUME_REDUCED over VOLUME_DOWN_TIME. While the translator keeps speaking the fade is sent again at most once per VOLUME_DOWN_TIME.
T2. The translator stops speaking.
T3. After SILENCE_LIMIT of silence the volume comes up to VOLUME_FULL, and after SILENCE_LIMIT2 it comes up again to VOLUME_FULL2. Raising it in two steps is smoother than one jump, and it fits a translation where the interpreter is almost always speaking, so the program never returns to full volume between sentences.

The volume is never raised before the translator has spoken for the first time since the script started. This keeps it from changing the bus when you start it.

Tuning

If the program also contains music or other parts that should be at full volume, set VOLUME_FULL2 to 100, so the program returns to full after a longer silence.
If the script reacts to breathing or room noise, raise VOICE_THRESHOLD. If it misses quiet speech, lower it. The script reads the left channel of the input's meter.
VOLUME_REDUCED is a percentage on vMix's volume scale, which is not linear. 50 on the scale is much quieter than half the loudness, see Audio levels.

Log messages

The script only prints its start-up lines, a missing-input notice, and Translator | Unexpected error: ... when a read fails, so it stays quiet while it works.