> ## Documentation Index
> Fetch the complete documentation index at: https://docs.firetone.com.au/llms.txt
> Use this file to discover all available pages before exploring further.

# The audio library

> Every recording the organisation has: prompts a flow plays, and the music a caller hears while waiting.

**Audio** is where an organisation's recordings live. Five things play from
it — a flow's prompts, a queue's hold music, the organisation's default hold
music, a campaign's message, and a virtual agent's music — which is why it is
a screen of its own rather than a tab on the flow designer. Filing it under
any one of the five would make it look like that one owned it.

Recordings belong to one organisation. The switch will not play another
organisation's audio, so with "All organisations" selected there is nothing
to show and nothing to upload until you pick one.

## Two libraries, not one

Every recording is either a **prompt** or **hold music**, and the runtime
keeps them apart: hold music is offered where hold music is chosen, prompts
where a greeting is. You choose which you are making when you add it.

| Kind           | Offered on                                                                |
| -------------- | ------------------------------------------------------------------------- |
| **Prompt**     | flow nodes, voicemail greetings, campaign messages                        |
| **Hold music** | queues, a flow's hold node, the organisation's default, a virtual agent's |

## Three ways to add one

<Steps>
  <Step title="Upload a file">
    The usual way. A WAV or MP3 of what you want played.
  </Step>

  <Step title="Generate from text">
    Type the words and have them spoken. Useful for a prompt that will change
    again next week, and for hearing how a wording sounds before somebody
    books a studio.
  </Step>

  <Step title="Record by phone">
    The panel rings an extension. Answer it, key the PIN, speak after the
    tone. This is how you get a recording in the voice of the person who
    answers the phones, with no file to move about.
  </Step>
</Steps>

## What a caller hears while waiting

Hold music is resolved **most specific first**, and the first level that has
something wins:

1. the **queue's** own music, or a flow hold node's;
2. the **organisation's** default;
3. the music that ships with the switch.

The last of those is real, not a placeholder: an organisation that configures
nothing still gets music. And because the organisation's default is only
looked up when nothing more specific answered, a queue with its own music
costs nothing extra to play.

## Deleting one

A recording that something still plays cannot be deleted, and the refusal
says what is still using it: an organisation's hold music or refusal
announcement, a queue's hold music, a virtual agent's, or an IVR flow node —
on **any revision**, published or draft, because a published revision is live
and a draft is about to be.

<Note>
  That is a better answer than a broken prompt. A flow whose greeting has
  been deleted plays silence at the moment a caller is deciding whether
  anybody is there.
</Note>

Reading the library needs `ivr:read`; adding and deleting need `ivr:write` —
the same permissions the flow designer has always used, because this is the
same audio.