AI4CALL assistant settings: what advanced management controls

The AI4CALL assistant settings are the advanced management section that controls conversation timing, interruptions and what happens during silence. The values you see on screen belong to this specific assistant: there is no single set of numbers that works for everyone, only the one that matches the calls you actually have to handle.

In the video tutorial, each parameter is shown where it really matters: when the assistant understands the caller has finished speaking, when it decides to end the line, when it nudges for an answer. Below you'll find the same sequence in written form, handy to keep open while you configure the panel.

End of speech and inactivity timeout: two silences not to confuse

End of speech defines how much silence to wait for before considering the caller's turn finished. A short wait speeds up the reply, but risks reading a pause as the end of a sentence. A longer wait gives the speaker more room and stretches the timing. Tune this balance on real sentences, pauses and second thoughts included.

The inactivity timeout deals instead with prolonged silence: if nobody speaks and no tools are running, the call is closed when the configured time expires. It is a different control from the short pause used to recognise the end of a sentence.

Speech recognition: start delay, truncate audio and voice activity detection

The recognition start delay sets when listening begins after the call starts: useful for handling the opening audio. The Truncate audio option discards the audio that precedes the start of recognition, so enable it only if you really want to throw that part away.

Voice activity detection tells speech apart from noise and keeps a noise from being read as the caller stepping in. Test it in less quiet environments too: line and microphone quality affect the result.

• End of speech: the silence that closes the caller's turn.
• Inactivity timeout: the prolonged silence that ends the call.
• Recognition start delay: when the assistant starts listening.
• Truncate audio: discards the audio before recognition starts.
• Voice activity detection: tells speech apart from noise.
• Report interruptions to the AI: informs the model when the caller interrupts.
• Maximum length: character limit on the text sent to speech synthesis.

Interruptions and maximum reply length

Report interruptions to the AI tells the model when the caller cuts into the reply. The assistant can then take into account that part of the message may not have been heard, and pick the conversation back up without treating as delivered information the customer never received.

The maximum length limits the text sent to speech synthesis: it is a character limit, separate from the model's token limit. On the phone, short answers are easier to follow, but still tell the assistant how to be concise in the prompt. A technical cap doesn't replace well-written instructions.

Reprompt and fixed phrases: how to nudge for an answer

The reprompt lets the assistant nudge for an answer when the caller stays silent. You can choose text generated by the model, a fixed phrase or the last sentence. The number of attempts limits how many times it retries: keep it low, to avoid an insistent string of nudges.

With fixed phrases, open Configure fixed phrases. The screen lets you prepare several messages, which are then picked at random: every message therefore has to work on its own.

For example: are you still on the line? Can I help you with anything else? Write a short sentence and add it to the list. Avoid references to a first or a second attempt, because the playback order is random.

Permissions: anonymous, international, mobile and landline numbers

In the permissions section you'll find the controls for anonymous, international, mobile and landline numbers, with the option to specify allowed prefixes. Check that they match the numbers the assistant is meant to handle, then apply the panel settings and save the assistant once the configuration is complete.

Frequently asked questions about assistant settings

What's the difference between end of speech and inactivity timeout?

End of speech is the short silence after which the assistant considers the caller's turn finished. The inactivity timeout deals with prolonged silence: if nobody speaks and no tools are running, the call is closed when the configured time expires.

What is the reprompt for?

The reprompt lets the assistant nudge for an answer when the caller stays silent. It can use text generated by the model, a fixed phrase or the last sentence, with a configurable number of attempts.

Can I copy the values shown in the video?

No: the visible values belong to that assistant and should be assessed against the calls you want to handle. In particular, the maximum length value on that screen does not match the range indicated by the interface and must not be used as a recommended value.