Speech customization

Configure speech settings that are not available as standard options by using bot parameters. Add or edit these parameters in the JSON editor on the bot connection's Advanced tab. See Manage bot connections. For example:

{
  "continuousASR": true,
  "continuousASRTimeoutInMS": 5000,
  "continuousASRDigits": "#"
}

Use continuous automatic speech recognition (ASR) to collect longer responses

By default, the speech-to-text service determines when an utterance ends, typically by detecting a pause in speech. Live Hub sends each recognized utterance to the bot as a separate message.

This behavior can interrupt callers who respond at length. If a caller speaks across multiple sentences, the speech-to-text service can split the response into several utterances, which the bot receives one at a time.

Enable continuous automatic speech recognition (ASR) to collect these utterances into a single response. Live Hub concatenates the recognized speech and sends it to the bot as one message when either of the following conditions is met:

• The caller has been silent for a specified period.
• The caller presses a designated DTMF key.

The bot can also enable, disable, or update continuous ASR settings during the conversation.

Parameter	Type	Description
continuousASR	boolean	Turns the feature on. Default false.
continuousASRTimeoutInMS	number	Silence, in milliseconds, that ends the collection and sends what has accumulated. The timer restarts on each recognition received from the speech-to-text service. From 500 to 60000, default 3000. Must be less than or equal to userNoInputTimeoutMS.
continuousASRDigits	string	DTMF key that sends the accumulated recognitions immediately. Default #.
continuousASRHypothesisTimeoutInMS	number	Timeout, in milliseconds, between hypotheses and between the last hypothesis and the recognition. When it expires, the caller's last utterance is discarded and the earlier recognitions go to the bot. From 500 to 6000, default 3000.
continuousASREmptyUserInputText	string	What to send the bot when the caller presses the end digit without saying anything. Default <empty>, literally. Set it to an empty string to send nothing at all.

Tune continuousASRTimeoutInMS to match your callers' behavior. A shorter timeout reduces delay, while a longer timeout gives callers more time to complete longer responses before Live Hub sends them to the bot.

Set continuousASREmptyUserInputText if you want the bot to detect callers who indicate that they have finished without saying anything. When this occurs, Live Hub sends the configured text to the bot, allowing it to prompt the caller again instead of continuing to wait for input.

Reduce background noise and improve interruption handling

Live Hub supports two audio processing capabilities designed to improve the quality and responsiveness of voice interactions.

• Krisp Noise Cancellation provides advanced background noise removal and speaker separation capabilities, reducing the impact of static noise, environmental sounds, and nearby conversations.
• Silero Voice Activity Detection (VAD) enhances speech activity detection, enabling faster identification of genuine user interruptions while reducing false interruption events caused by background noise and other non-speech audio. This creates more natural conversations and improves turn-taking between users and AI agents.

Together, these enhancements deliver a better voice experience by increasing transcription accuracy, improving turn detection, and enabling faster and more accurate interruption handling.

Availability: These capabilities are currently available as part of a limited trial.

• Can only be enabled by an AudioCodes administrator.
• Are currently supported only in the Frankfurt region.
• Require assistance from AudioCodes Support for activation.

Customers interested in evaluating these capabilities should contact AudioCodes Support for onboarding and trial enablement.