Adaptive voicemail and IVR detection for AI-driven call automation
The present disclosure provides a system for adaptive voicemail and interactive voice response (IVR) detection in outbound calls. The system includes a call initialization module configured to establish an outbound call connection, a speech processing module configured to convert incoming audio signals into text in real-time, a classification module configured to analyze the text and determine whether the call has reached a live recipient, a voicemail system, or an IVR menu, a decision-making module configured to determine an appropriate course of action based on the classification, and a response generation module configured to generate and deliver appropriate responses based on the determined course of action. The system enables efficient handling of outbound calls by accurately detecting and responding to different call scenarios.
1 . A computer-implemented method for adaptive outbound call handling, the method comprising:
capturing, with at least one processor, audio of a voice call established between a calling system and a called endpoint and dividing the audio into successive frames;
assigning, by the at least one processor, a speech-recognition confidence value to each frame and discarding frames having confidence values below a configurable threshold prior to further processing;
verifying, by the at least one processor, that a transmission queue is free of system-generated outbound audio and, upon that verification, converting remaining frames into text and classifying the voice call as voicemail, interactive voice response, or live human speech by executing a machine-learning language model on the text;
selecting, by the at least one processor in dependence on the classification, a corresponding call-handling action;
when the classification identifies the interactive voice-response state, generating, by the at least one processor and in real time, control inputs derived from a stored menu-navigation policy that at least maps recognised menu wording to dual-tone-multi-frequency sequences, updates the mapping in response to menu transitions detected in the inbound audio, and continues to issue updated control inputs until detection of live-speaker pickup; and
executing, by the at least one processor, the selected call-handling action, the call-handling action further comprising delivering a prerecorded voicemail message when the classification identifies the voicemail state and continuing conversational interaction when the classification identifies the live-speaker state.
2 . The method of claim 1 , wherein the configurable threshold for discarding frames is adjustable based on ambient noise levels detected in the audio.
3 . The method of claim 1 , wherein converting remaining frames into text comprises using a speech-to-text engine trained on telephony audio data.
4 . The method of claim 1 , wherein the machine-learning language model is a transformer-based model fine-tuned on labeled examples of voicemail greetings, IVR prompts, and live human speech.
5 . The method of claim 1 , further comprising logging call outcomes and updating the stored menu-navigation policy based on successful navigation patterns.
6 . The method of claim 1 , wherein the control inputs for interactive voice response navigation include DTMF tones and synthesized speech responses.
7 . The method of claim 1 , further comprising detecting silence periods in the audio and using the detected silence periods to determine appropriate timing for delivering the prerecorded voicemail message.
8 . The method of claim 1 , wherein continuing conversational interaction comprises activating a natural language processing module to engage in dialogue with the live speaker.
9 . The method of claim 1 , further comprising updating the classification in real-time as the call progresses, allowing for dynamic switching between call-handling actions.
10 . A system for adaptive outbound call handling, comprising:
at least one processor; and
a memory storing instructions that, when executed by the at least one processor, cause the system to:
capture audio of a voice call established between a calling system and a called endpoint and divide the audio into successive frames;
assign a speech-recognition confidence value to each frame and discard frames having confidence values below a configurable threshold prior to further processing;
verify that a transmission queue is free of system-generated outbound audio and, upon that verification, convert remaining frames into text and classify the voice call as voicemail, interactive voice response, or live human speech by executing a machine-learning language model on the text;
select, in dependence on the classification, a corresponding call-handling action;
when the classification identifies the interactive voice-response state, generate, in real time, control inputs derived from a stored menu-navigation policy that at least maps recognised menu wording to dual-tone-multi-frequency sequences, updates the mapping in response to menu transitions detected in the inbound audio, and continues to issue updated control inputs until detection of live-speaker pickup; and
execute the selected call-handling action, the call-handling action further comprising delivering a prerecorded voicemail message when the classification identifies the voicemail state and continuing conversational interaction when the classification identifies the live-speaker state.
11 . The system of claim 10 , further comprising a telephony interface for establishing the voice call with the called endpoint.
12 . The system of claim 10 , wherein the memory further stores a database of prerecorded voicemail messages selectable based on call context.
13 . The system of claim 10 , further comprising a user interface for configuring the stored menu-navigation policy.
14 . A non-transitory computer-readable storage medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to:
capture audio of a voice call established between a calling system and a called endpoint and divide the audio into successive frames;
assign a speech-recognition confidence value to each frame and discard frames having confidence values below a configurable threshold prior to further processing;
verify that a transmission queue is free of system-generated outbound audio and, upon that verification, convert remaining frames into text and classify the voice call as voicemail, interactive voice response, or live human speech by executing a machine-learning language model on the text;
select, in dependence on the classification, a corresponding call-handling action;
when the classification identifies the interactive voice-response state, generate, in real time, control inputs derived from a stored menu-navigation policy that at least maps recognised menu wording to dual-tone-multi-frequency sequences, updates the mapping in response to menu transitions detected in the inbound audio, and continues to issue updated control inputs until detection of live-speaker pickup; and
execute the selected call-handling action, the call-handling action further comprising delivering a prerecorded voicemail message when the classification identifies the voicemail state and continuing conversational interaction when the classification identifies the live-speaker state.
15 . The non-transitory computer-readable storage medium of claim 14 , wherein the instructions further cause the computing system to perform real-time analytics on call outcomes and adjust call-handling strategies based on the analytics.