Designing the sound of a Space that has a reading bot in it
A reading bot puts human voices, synthesised speech and sound effects in the same room. Why the effects get worse as the room fills, and how to cut the noise without losing the reading.
Adding a reading bot to a Space adds kinds of sound. The hosts and speakers, the synthesised voice reading comments, and the sound effects from reactions. Three things in one place.
When it works you do not think about it. When it does not, it arrives as a single word: loud. Splitting up what is actually loud changes what you do about it.
It is almost always the effects
Among people using Comeyomi, complaints about sound land almost entirely on the effects.
The reaction sounds are loud.
Once there are people and the reactions pile up, it gets really noisy — first thing I'd drop.
Meanwhile, complaints about the reading voice itself, or the reading speed, barely appear. That may be counterintuitive: told that "the reading is loud", the instinct is to turn the reading off.
The reason holds up. Speech plays once per comment and runs until the sentence ends. Its length is set by the text, and two do not run at once.
Effects are different. A reaction takes an instant to send, and any number of people can send at once. The rate climbs with the number of people, and each sound is short and sharp, so it cuts through whatever is being said.
A setting that felt right at ten listeners does not hold at fifty. Comments grow roughly in proportion to the room; reactions grow with how effortless they are.
There is more than one volume
The useful thing to know: speech volume and effect volume are separate settings.
Turning the speech down does not make the effects quieter, and the reverse is true too. If lowering "the volume" did not help, you lowered the wrong one.
In Comeyomi, the chat settings carry an effect volume from 0% to 200%, defaulting to 70%.
So you can keep the reading and cut the effects. When the room fills up, that is the first move. There is no need to stop reading comments: the reading relates to the point of the Space, while the effects are decoration.
Three positions
How much sound a segment needs varies. Comeyomi's three reading modes line up with that.
| Segment | Mode | Effects |
|---|---|---|
| Taking questions and reactions | Full | Adjust to taste |
| Hosts talking among themselves | SE | Reactions still audible |
| Presenting, explaining | Off | Silent |
"SE" plays a tone and the effects without speaking the text. It is not silence. It suits wanting to know a reaction arrived without stopping the conversation.
For quiet, choose "Off". Picking SE as a quiet mode gets you the opposite of what you meant.
Say when you switch. Turning it off silently leaves listeners unable to tell whether their comments failed to arrive or simply are not being read.
Timing is not a volume problem
The other recurring complaint is about timing rather than level.
There's lag, and sometimes it just starts reading over you.
It can read comments, but it can't read the room.
When a synthesised voice lands on top of a person, neither is intelligible. Lowering the volume does not fix an overlap.
The conversation wait covers this: while somebody is speaking, the bot holds the reading and starts once the room falls quiet. On, off, or automatic.
Automatic switches according to how busy the Space is — reading straight away when it is calm, waiting once it fills up.
Comments still queue while it waits, though, and when things are busy, older queued ones may be dropped. Do not build a format that assumes every comment is read.
Putting it together by size
Up to ten or so. The defaults mostly hold. The effects work as atmosphere. Automatic is enough for the conversation wait.
Tens of people. Time to start lowering the effect volume. If reactions are landing on speech, 40–50% is a reasonable place to look. Set the conversation wait to on so the hosts do not get talked over.
Large, or presentation-shaped. Cut the effects further, or switch to SE and stop reading the text. Present in Off and return to Full for questions.
The numbers are a starting point. Microphones, headphones and how loudly people speak all move them, so send a short message and listen.
Before you start
When introducing a bot to a Space, this order runs into the fewest surprises.
- Set the mode to Full and send a short message to confirm the reading
- Send one reaction and hear the effect volume for yourself
- If it is loud, lower the effect volume then and there
- As people arrive, turn the conversation wait on
Step two is the one people skip, and it is the one that causes trouble later. Noticing after the broadcast has started means asking the audience to put up with it.
The settings are described in reading settings, and which symbol plays which sound is in sound effects.