{"id":2627,"date":"2026-11-19T12:38:54","date_gmt":"2026-11-19T12:38:54","guid":{"rendered":"https:\/\/www.capanicus.com\/blog\/?p=2627"},"modified":"2026-11-19T12:38:59","modified_gmt":"2026-11-19T12:38:59","slug":"ai-answering-machine-detection-for-asterisk-dialers","status":"publish","type":"post","link":"https:\/\/www.capanicus.com\/blog\/ai-answering-machine-detection-for-asterisk-dialers\/","title":{"rendered":"AI Answering Machine Detection for Asterisk Dialers"},"content":{"rendered":"<p><span style=\"font-weight: 400;\">Every outbound team knows the cost of a wrong answering machine call. The dialer connects an agent to a voicemail greeting, or worse, hangs up on a real person who just said &#8220;Hello?&#8221;. On Asterisk, the built-in AMD() application has handled this job for years, but its timing rules struggle with today&#8217;s voicemail greetings, mobile networks and noisy lines. This guide explains how AI answering machine detection works and how it compares with Asterisk&#8217;s AMD(). It then shows how to add it to an Asterisk dialer, how to measure and tune accuracy, and which FCC abandoned-call rules your settings must respect.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">It is written for teams that run or build Asterisk-based dialers: call centers, BPOs, VoIP providers and product teams. Each section ends with something you can act on, from a dialplan example to a test plan.<\/span><\/p>\n<div style=\"background:#fff8f1;border:1px solid #fde3cc;border-left:5px solid #fb8500;border-radius:10px;padding:20px 22px;margin:24px 0;\">\n<p style=\"margin:0 0 14px;font-size:18px;font-weight:700;color:#001d26;\">Quick Summary<\/p>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 12px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">1<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">Answering machine detection (AMD) decides in the first seconds whether a person or a machine answered, then routes the call.<\/span><\/div>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 12px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">2<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">Asterisk&#8217;s AMD() uses silence and word-timing rules from amd.conf. It is free, but short or formal greetings can fool it.<\/span><\/div>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 12px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">3<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">AI answering machine detection classifies the audio itself, using acoustic models, speech-to-text, or both.<\/span><\/div>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 12px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">4<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">You can add AI AMD to Asterisk by streaming early call audio over AudioSocket, ARI External Media or EAGI.<\/span><\/div>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 12px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">5<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">Tune for false positives first, because hanging up on a live person hurts customers and compliance.<\/span><\/div>\n<div style=\"display:flex;align-items:flex-start;gap:12px;margin:0 0 0px;\"><span style=\"flex:0 0 28px;width:28px;height:28px;border-radius:50%;background:#001d26;color:#ffffff;font-size:14px;font-weight:700;line-height:28px;text-align:center;\">6<\/span><span style=\"font-size:16px;line-height:1.6;color:#102c36;\">In the US, abandoned calls must stay at or below 3% of live-answered calls per campaign over 30 days.<\/span><\/div>\n<\/div>\n<h2><b>What Answering Machine Detection Does in an Outbound Dialer<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">Answering machine detection (AMD) is the step that listens to the first seconds of an answered outbound call and decides who, or what, picked up. The verdict tells the dialer what to do next: connect an agent, leave a voicemail, retry later or hang up.<\/span><\/p>\n<h3><b>The Decision AMD Makes in the First Few Seconds<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">When a call is answered, the carrier sends an answer signal. That signal looks the same whether a person, a voicemail box, an IVR or a fax machine picked up. AMD has to tell them apart from the audio alone.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Most systems return one of three results: human, machine or not sure. Some also flag fax tones, carrier announcements, silence and IVR menus, so the dialer can handle each one differently.<\/span><\/p>\n<h3><b>Where AMD Fits in Predictive, Progressive and Power Dialing<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">A predictive dialer places more calls than there are free agents, based on expected answer rates. AMD matters most here, because the dialer wants only live people to reach agents. That is why predictive dialer AMD settings get the most attention in any outbound setup. If you are new to pacing, this explainer on<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/predictive-dialer-and-how-it-increase-call-center-efficiency\/\"><span style=\"font-weight: 400;\">how a predictive dialer works<\/span><\/a><span style=\"font-weight: 400;\"> covers the basics.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">In progressive and power modes, the dialer calls only when an agent is free, so AMD saves agent time rather than protecting pacing. In preview mode, the agent places the call and hears the answer directly, so AMD is usually switched off.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Because pacing, agent idle time and abandoned calls all depend on it, accurate detection is a core part of<\/span> <a href=\"https:\/\/www.capanicus.com\/predictive-dialer-development\"><span style=\"font-weight: 400;\">custom predictive dialer development<\/span><\/a><span style=\"font-weight: 400;\">.<\/span><\/p>\n<h3><b>Why Voicemail Is Harder to Spot Than It Used to Be<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Old answering machines played long greetings and ended with a clear beep. Today, outbound calls reach mobile voicemail and carrier messages. They also reach call-screening assistants on modern phones and short personal greetings like &#8220;Hi, it&#8217;s Sam.&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Some of these greetings sound almost exactly like a live person. Network delay, comfort noise and compressed codecs also blur the silence gaps that simple voicemail detection depends on. That is the gap AI answering machine detection tries to close.<\/span><\/p>\n<h2><b>How Asterisk&#8217;s Built-In AMD() Application Decides<\/b><\/h2>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-2644\" src=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/Asterisk-AMD-Decision-Flow.webp\" alt=\"How Asterisk AMD() decides between human, machine and not sure\" width=\"1000\" height=\"660\" srcset=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/Asterisk-AMD-Decision-Flow.webp 1000w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/Asterisk-AMD-Decision-Flow-300x198.webp 300w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/Asterisk-AMD-Decision-Flow-768x507.webp 768w\" sizes=\"auto, (max-width: 1000px) 100vw, 1000px\" \/><\/p>\n<p><span style=\"font-weight: 400;\">Asterisk ships with an AMD() dialplan application in the <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">app_amd<\/code> module, and it has been available since Asterisk 1.4. It is free and runs inside Asterisk. It does not understand words, though. It measures voice and silence, then applies timing rules.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">If you are still choosing a platform for a new dialer, our<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/freeswitch-vs-asterisk\/\"><span style=\"font-weight: 400;\">FreeSWITCH vs Asterisk<\/span><\/a><span style=\"font-weight: 400;\"> comparison covers the bigger trade-offs.<\/span><\/p>\n<h3><b>The amd.conf Settings That Control Detection<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">AMD() reads its defaults from <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">amd.conf<\/code> when the module loads, and any arguments you pass in the dialplan override them. All durations are in milliseconds. The sample <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">amd.conf<\/code> that ships with current Asterisk uses these values:<\/span><\/p>\n<table>\n<tbody>\n<tr>\n<td><b>Setting<\/b><\/td>\n<td><b>Sample value<\/b><\/td>\n<td><b>What it controls<\/b><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">initial_silence<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">2500<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Longest silence allowed before the greeting starts. Exceeded: MACHINE<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">greeting<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">1500<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Longest greeting allowed. Exceeded: MACHINE<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">after_greeting_silence<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">800<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Silence after a greeting that signals a person waiting for a reply. Exceeded: HUMAN<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">total_analysis_time<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">5000<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Maximum time AMD may take to decide<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">min_word_length<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">100<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Shortest burst of voice counted as a word<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">between_words_silence<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">50<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Silence needed before the next sound counts as a new word<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_number_of_words<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">2<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Most words allowed in a greeting. Exceeded: MACHINE<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_word_length<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">5000<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Longest burst of continuous voice accepted as one word. Exceeded: MACHINE<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">silence_threshold<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">256<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Average noise level (0 to 32767) below which audio counts as silence<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Older Asterisk releases shipped different sample values, so check the <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">amd.conf<\/code> on your own server before you tune anything.<\/span><\/p>\n<h3><b>Reading AMDSTATUS and AMDCAUSE<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">When AMD() finishes, it sets two channel variables. <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDSTATUS<\/code> holds the verdict: MACHINE, HUMAN, NOTSURE or HANGUP. <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> explains the verdict with a cause name followed by the measured values:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>INITIALSILENCE:<\/b><span style=\"font-weight: 400;\"> nobody spoke before <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">initial_silence<\/code> ran out, so the call is marked MACHINE.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>LONGGREETING:<\/b><span style=\"font-weight: 400;\"> the greeting ran longer than the <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">greeting<\/code> setting allows.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>MAXWORDS:<\/b><span style=\"font-weight: 400;\"> the greeting had more words than <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_number_of_words<\/code>.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>MAXWORDLENGTH:<\/b><span style=\"font-weight: 400;\"> one burst of voice ran longer than <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_word_length<\/code>.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>HUMAN:<\/b><span style=\"font-weight: 400;\"> a short greeting was followed by enough silence, so the call is marked HUMAN.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>TOOLONG:<\/b><span style=\"font-weight: 400;\"> no decision was reached within <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">total_analysis_time<\/code>, so the result is NOTSURE.<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">Log both variables on every call. <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> is the fastest way to see which rule is behind a wrong result.<\/span><\/p>\n<h3><b>A Basic AMD Dialplan Example<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Run AMD() only after the outbound call is answered. This example sends live answers and uncertain results to an agent queue, and plays a short message to answering machines:<\/span><\/p>\n<div style=\"margin:16px 0;\">\n<div style=\"background:#001d26;color:#ffffff;font-family:Consolas,Monaco,'Courier New',monospace;font-size:13px;padding:6px 16px;border-radius:6px 6px 0 0;\">extensions.conf<\/div>\n<pre style=\"background:#f6f8fa;border:1px solid #e1e4e8;border-top:none;border-left:4px solid #fb8500;border-radius:0 0 6px 6px;padding:14px 16px;margin:0;overflow-x:auto;white-space:pre;font-family:Consolas,Monaco,'Courier New',monospace;font-size:14px;line-height:1.55;color:#24292e;\"><code><span style=\"color:#6f42c1;font-weight:600;\">[dialer-amd]<\/span>\n<span style=\"color:#d73a49;\">exten<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> s,1,<span style=\"color:#005cc5;\">NoOp<\/span>(Outbound call answered)\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">AMD<\/span>()\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">NoOp<\/span>(AMD result: <span style=\"color:#e36209;\">${AMDSTATUS}<\/span> \/ <span style=\"color:#e36209;\">${AMDCAUSE}<\/span>)\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">GotoIf<\/span>($[\"<span style=\"color:#e36209;\">${AMDSTATUS}<\/span>\" = <span style=\"color:#032f62;\">\"MACHINE\"<\/span>]?machine,1)\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">Queue<\/span>(sales-agents) \u00a0 <span style=\"color:#6a737d;font-style:italic;\">; HUMAN and NOTSURE go to agents<\/span>\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">Hangup<\/span>()\n\n<span style=\"color:#d73a49;\">exten<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> machine,1,<span style=\"color:#005cc5;\">WaitForSilence<\/span>(1500,1,20)\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">Playback<\/span>(custom\/voicemail-message)\n\u00a0<span style=\"color:#d73a49;\">same<\/span> <span style=\"color:#d73a49;\">=&gt;<\/span> n,<span style=\"color:#005cc5;\">Hangup<\/span>()<\/code><\/pre>\n<\/div>\n<p><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">WaitForSilence()<\/code> waits for the greeting to go quiet before the message plays. It is a simple stand-in for beep detection, so test it against real greetings from your own call lists. Leaving prerecorded messages also has consent rules, covered in the compliance section below.<\/span><\/p>\n<h3><b>Where Rule-Based Detection Breaks Down<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Timing rules work well on classic greetings and poorly on everything else. These are the patterns that cause most wrong results:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">A person who answers and waits silently can run past <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">initial_silence<\/code> and be marked MACHINE.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">A person who answers with a full sentence, such as &#8220;Hello, this is Priya from accounts,&#8221; can exceed <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_number_of_words<\/code> and be marked MACHINE.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">A short voicemail greeting such as &#8220;Hi, it&#8217;s Sam,&#8221; followed by a pause, can look like a person and reach an agent.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Background noise and comfort noise can hide silence, which confuses every timing rule.<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">Tuning one rule to fix one error often creates another. Raising <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_number_of_words<\/code> means fewer live people get marked MACHINE by mistake, but more voicemail reaches agents. This is one of the<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/asterisk-development-challenges\/\"><span style=\"font-weight: 400;\">Asterisk development challenges<\/span><\/a><span style=\"font-weight: 400;\"> that dialer teams meet as volume grows, and it is why many move to AI answering machine detection.<\/span><\/p>\n<h2><b>How AI Answering Machine Detection Works<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">AI answering machine detection replaces fixed timing rules with a model trained on real call audio. Instead of counting words and pauses, the model learns what live answers, voicemail greetings, IVRs and carrier messages sound like. Most systems use one of the three approaches below, or a mix of them.<\/span><\/p>\n<h3><b>Acoustic Models That Classify the Audio Directly<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">An acoustic model turns the first second or two of audio into features, such as a spectrogram, and classifies them. It can pick up cues that timing rules miss. Examples include the flat, compressed sound of a recorded greeting, the background sound of a live phone and the steady rhythm of a scripted message.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Because it works on sound rather than words, an acoustic model can decide quickly. It can also work across languages, as long as its training data covers the languages, accents and carriers you call.<\/span><\/p>\n<h3><b>Speech-to-Text Models That Read the Greeting<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">A second approach transcribes the greeting with streaming speech-to-text and checks the words. Phrases like &#8220;you&#8217;ve reached&#8221;, &#8220;not available&#8221; or &#8220;leave a message after the tone&#8221; are strong machine signals. A short &#8220;Hello?&#8221; followed by silence is a strong human signal.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">This approach is easy to explain and audit, because you can read why a call was marked MACHINE. It does add transcription delay, though, and it needs speech-to-text support for every language you dial.<\/span><\/p>\n<h3><b>Beep and Tone Detection for Voicemail Drop<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">If your campaign leaves voicemail, spotting the machine is only half the job. The message has to start after the beep, or the first part of it is lost.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">AI systems often add a separate step that listens for the end of the greeting and the beep. The same step can catch fax and modem tones and the special information tones that networks play for disconnected or unreachable numbers.<\/span><\/p>\n<h3><b>Rule-Based AMD vs AI AMD at a Glance<\/b><\/h3>\n<table>\n<tbody>\n<tr>\n<td><b>Factor<\/b><\/td>\n<td><b>Asterisk AMD()<\/b><\/td>\n<td><b>AI answering machine detection<\/b><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">How it decides<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Silence and word-timing rules<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A model trained on labeled call audio<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Setup<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Built in, configured in <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">amd.conf<\/code><\/span><\/td>\n<td><span style=\"font-weight: 400;\">A separate service connected to Asterisk<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Cost<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Free<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Hosting and compute, or per-call API fees<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Short or formal greetings<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Often misclassified<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Handled better when trained on similar calls<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Languages and accents<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Not language-aware<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Depends on the training data<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Tuning<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Manual threshold changes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Confidence thresholds and retraining<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Explaining a result<\/span><\/td>\n<td><span style=\"font-weight: 400;\"><code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> shows the rule that fired<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Needs confidence scores and logs<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Best fit<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Low volume, simple campaigns<\/span><\/td>\n<td><span style=\"font-weight: 400;\">High volume, mobile-heavy or multi-country campaigns<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Many teams keep AMD() as a fallback. If the AI service is slow or unreachable, the call still gets a verdict.<\/span><\/p>\n<h2><b>How to Add AI AMD to an Asterisk Dialer, Step by Step<\/b><\/h2>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-2643\" src=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AI-AMD-Asterisk-Architecture.webp\" alt=\"AI AMD architecture where Asterisk streams call audio to a detection service and routes the call\" width=\"1000\" height=\"624\" srcset=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AI-AMD-Asterisk-Architecture.webp 1000w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AI-AMD-Asterisk-Architecture-300x187.webp 300w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AI-AMD-Asterisk-Architecture-768x479.webp 768w\" sizes=\"auto, (max-width: 1000px) 100vw, 1000px\" \/><\/p>\n<p><span style=\"font-weight: 400;\">You do not need to replace Asterisk to use AI answering machine detection. The usual pattern keeps Asterisk in charge of the call and sends the first seconds of audio to a separate detection service. The diagram shows the flow, and the four steps below explain each part.<\/span><\/p>\n<h3><b>Step 1: Stream the First Seconds of Call Audio Out of Asterisk<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Asterisk offers three common ways to get live call audio to an outside service:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>AudioSocket:<\/b><span style=\"font-weight: 400;\"> a simple two-way audio protocol over TCP, added to Asterisk in 2020 as a dialplan application and a channel driver.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>ARI External Media:<\/b><span style=\"font-weight: 400;\"> an ARI request that creates a channel streaming the call&#8217;s audio to an external host over RTP or AudioSocket. Pair it with a snoop channel to listen without changing the call.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>EAGI:<\/b><span style=\"font-weight: 400;\"> an AGI script that receives the caller&#8217;s audio on an extra file descriptor. It is simple to start with, but it runs one process per call.<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">Stream only what the model needs, usually signed linear audio at 8 or 16 kHz, from the moment of answer until a verdict comes back. This is the same audio path teams use to connect<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/integrate-ai-voice-agents-pbx\/\"><span style=\"font-weight: 400;\">AI voice agents to a PBX<\/span><\/a><span style=\"font-weight: 400;\">, so one integration can serve both jobs.<\/span><\/p>\n<h3><b>Step 2: Classify the Audio in Real Time<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">The detection service reads short audio frames and updates its prediction as each one arrives. It should return a verdict as soon as confidence passes your threshold, and never later than a hard time limit you set.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Typical outputs are human, machine, not sure, fax and silence, each with a confidence score. Keep the service close to Asterisk on the network, because every round trip adds dead air for the person on the line.<\/span><\/p>\n<h3><b>Step 3: Return the Verdict and Route the Call<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">The service sends its result back to Asterisk, for example by setting a channel variable over ARI or by returning it to an AGI script. The dialplan then routes the call the same way it routes <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDSTATUS<\/code>:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Human:<\/b><span style=\"font-weight: 400;\"> bridge the call to the next free agent, or hand it to a<\/span> <a href=\"https:\/\/www.capanicus.com\/ai-voice-agent-development\"><span style=\"font-weight: 400;\">custom AI voice agent<\/span><\/a><span style=\"font-weight: 400;\"> for qualification.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Machine:<\/b><span style=\"font-weight: 400;\"> hang up and schedule a retry, or wait for the beep and play a voicemail message.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Not sure:<\/b><span style=\"font-weight: 400;\"> send the call to an agent, because a few seconds of agent time cost less than hanging up on a customer.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Timeout or error:<\/b><span style=\"font-weight: 400;\"> fall back to AMD(), or route the call to an agent.<\/span><\/li>\n<\/ul>\n<h3><b>Step 4: Log Every Verdict and Compare It With Outcomes<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Store the verdict, confidence score, decision time and, if AMD() also ran, <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> for every call. Then compare those fields with what actually happened: agent dispositions, call length and, where recording is allowed, a sample of recordings.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Agent dispositions are your cheapest error report. When an agent marks a call as &#8220;answering machine&#8221; after AMD said human, you have found a missed machine. Sending these fields to your CRM through<\/span> <a href=\"https:\/\/www.capanicus.com\/contact-center-crm-integration\"><span style=\"font-weight: 400;\">contact center CRM integration<\/span><\/a><span style=\"font-weight: 400;\"> lets supervisors see detection quality next to campaign results.<\/span><\/p>\n<h2><b>Measuring Accuracy, Speed and Routing Decisions<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">An AMD setup is only as good as its most expensive mistake. Before you tune anything, agree on what each type of error costs your business.<\/span><\/p>\n<h3><b>False Positives vs False Negatives<\/b><\/h3>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-2642\" src=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AMD-False-Positive-vs-False-Negative.webp\" alt=\"AMD false positive versus false negative in an outbound dialer\" width=\"1000\" height=\"540\" srcset=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AMD-False-Positive-vs-False-Negative.webp 1000w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AMD-False-Positive-vs-False-Negative-300x162.webp 300w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/AMD-False-Positive-vs-False-Negative-768x415.webp 768w\" sizes=\"auto, (max-width: 1000px) 100vw, 1000px\" \/><\/p>\n<p><span style=\"font-weight: 400;\">A false positive marks a live person as a machine. The dialer hangs up on, or plays a recording to, someone who answered. That annoys the customer, and under US rules it can count as an abandoned call.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A false negative marks a machine as human. An agent hears a voicemail greeting and loses a few seconds. Whether you run AMD() or AI answering machine detection, most teams accept more false negatives to keep false positives low.<\/span><\/p>\n<h3><b>The Trade-Off Between Detection Time and Dead Air<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Every extra moment of analysis is silence for the person who answered. Many people say &#8220;Hello?&#8221; once or twice, hear nothing and hang up.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A detector that waits longer is usually more accurate, but it loses more live calls. It also uses up more of the two-second window that US rules allow after a greeting. Track the median and the slowest decision times, not only accuracy.<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/why-your-call-center-is-experiencing-high-latency\/\"><span style=\"font-weight: 400;\">Network latency in your call center<\/span><\/a><span style=\"font-weight: 400;\"> adds to this delay, so check the media path too.<\/span><\/p>\n<h3><b>Deciding What Happens to NOTSURE Results<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Every detector returns some uncertain results, so decide the rule per campaign. For sales and service calls, sending NOTSURE to an agent is the safer default. For high-volume reminder campaigns, retrying later can make more sense.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Watch how often NOTSURE appears. A rising share usually means a carrier, codec or greeting pattern has changed.<\/span><\/p>\n<h3><b>How to Test and Tune AMD Before Going Live<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">A careful rollout prevents most AMD surprises. Before you move a campaign to AI answering machine detection, use this order:<\/span><\/p>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Build a labeled test set from your own calls: live answers, mobile voicemail, business voicemail, IVRs, carrier messages and silence, for every country and language you call.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Run the new detector in shadow mode. It listens and logs, but your current AMD setup still routes the calls.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Compare false positives, false negatives, NOTSURE share and decision time between the two systems.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Switch one campaign at a time, starting with the campaign where false positives cost the most.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Re-test after any change to carriers, codecs, models or campaign scripts.<\/span><\/li>\n<\/ol>\n<h2><b>Compliance Rules That Shape Your AMD Settings<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">AMD settings are not only a performance choice. In the US, they decide whether a campaign stays inside federal telemarketing rules. This section is a summary, not legal advice, so confirm the rules for your campaigns with your legal team.<\/span><\/p>\n<h3><b>The FCC&#8217;s 3% Abandoned Call Limit<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Under the<\/span> <a href=\"https:\/\/www.law.cornell.edu\/cfr\/text\/47\/64.1200#:~:text=Abandon%20more%20than%20three%20percent%20of%20all%20telemarketing%20calls\"><span style=\"font-weight: 400;\">FCC rule<\/span><\/a><span style=\"font-weight: 400;\">, a telemarketer may not abandon more than three percent of telemarketing calls answered live by a person. The rate is measured over a 30-day period for a single calling campaign. Calls answered by a machine are not live answers, so they do not count toward that limit.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">This is where AMD accuracy becomes a compliance issue. Every live person who is hung up on or left in silence by mistake can push the abandoned rate up.<\/span><\/p>\n<h3><b>The Two-Second Rule After the Greeting<\/b><\/h3>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-2641\" src=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/FCC-Two-Second-Rule-AMD.webp\" alt=\"FCC two-second rule for connecting an agent after the called person\u2019s greeting\" width=\"1000\" height=\"620\" srcset=\"https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/FCC-Two-Second-Rule-AMD.webp 1000w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/FCC-Two-Second-Rule-AMD-300x186.webp 300w, https:\/\/www.capanicus.com\/blog\/wp-content\/uploads\/2026\/11\/FCC-Two-Second-Rule-AMD-768x476.webp 768w\" sizes=\"auto, (max-width: 1000px) 100vw, 1000px\" \/><\/p>\n<p><span style=\"font-weight: 400;\">A call counts as abandoned if it is not connected to a live sales representative within two seconds of the called person&#8217;s completed greeting. If no agent is free, the dialer must play a prerecorded identification and opt-out message within those two seconds.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Your detection time, your <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">after_greeting_silence<\/code> setting and your agent routing all have to fit inside this window. That is why slow detection is a compliance risk, not just an annoyance.<\/span><\/p>\n<h3><b>Ring Time, Voicemail Messages and Consent<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">The same FCC rule also says an unanswered telemarketing call must ring for at least 15 seconds or four rings before the dialer disconnects it. Set your no-answer timeout with this in mind.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Leaving prerecorded marketing messages on voicemail raises separate consent questions under the TCPA, so check them before you turn on voicemail drop. The FTC&#8217;s Telemarketing Sales Rule sets matching abandoned-call limits for the calls it covers.<\/span><\/p>\n<h3><b>Rules Outside the US<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Other markets have their own rules. In the UK, Ofcom treats abandoned and silent calls as persistent misuse. It also expects call centres that use AMD to include a reasoned estimate of AMD false positives in their abandoned call rate. In India, TRAI regulates commercial calls under its customer preference regulations.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">If you dial across borders, set your AMD and routing rules per country, not once for the whole platform.<\/span><\/p>\n<h2><b>Common AMD Problems on Asterisk and How to Fix Them<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">Most AMD complaints fall into a few patterns. Start with the symptom, check <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> or the AI confidence score, then apply the fix.<\/span><\/p>\n<h3><b>Live Callers Are Being Hung Up On<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">This is usually an <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">initial_silence<\/code> or <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">maximum_number_of_words<\/code> problem. People who pause before speaking, or answer with a full sentence, get marked MACHINE.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Pull the <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDCAUSE<\/code> values for the complaint calls first. Then raise the setting behind them in small steps, send NOTSURE to agents, and consider AI answering machine detection for the campaigns where this keeps happening.<\/span><\/p>\n<h3><b>Agents Keep Hearing Voicemail Greetings<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Short personal greetings are the usual cause. A greeting like &#8220;Hi, it&#8217;s Sam&#8221; followed by a pause passes the <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">after_greeting_silence<\/code> test and looks human.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Ask agents to tag these calls with an &#8220;answering machine&#8221; disposition, so you can measure the problem. A model trained on short greetings, or a transcript check for phrases like &#8220;leave a message&#8221;, catches many of them.<\/span><\/p>\n<h3><b>Long Silence After the Person Says Hello<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Dead air comes from three places: long analysis time, a slow round trip to the detection service, or no free agent when the verdict arrives.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Let the detector return early once it is confident, host it close to Asterisk, and check that dialer pacing matches real agent availability. Each fix shortens the gap the caller hears.<\/span><\/p>\n<h3><b>Results Change by Carrier, Codec or Country<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Carriers differ in early media, comfort noise and codecs, and voicemail platforms differ by country. A detector tuned on one route can drift on another.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Report detection results by carrier route and by country, not only overall. Test every new carrier in your<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/sip-trunking-architecture\/\"><span style=\"font-weight: 400;\">SIP trunking setup<\/span><\/a><span style=\"font-weight: 400;\"> before it carries campaign traffic.<\/span><\/p>\n<h2><b>When It Makes Sense to Build a Custom AI AMD Dialer<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">Not every team needs AI answering machine detection on day one. A small campaign with classic greetings can run well on a tuned AMD(). The signs below suggest your setup has outgrown it.<\/span><\/p>\n<h3><b>Signs Your Current Detection Setup Has Hit Its Limit<\/b><\/h3>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Customers complain about calls that hang up after they say hello.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Your abandoned rate sits close to the 3% limit even with careful pacing.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Agents spend a noticeable part of each hour on voicemail greetings.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">NOTSURE results keep growing, or results change when you add a carrier.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">You are expanding to new countries, languages or mostly mobile lists.<\/span><\/li>\n<\/ul>\n<h3><b>Build, Buy or Plug In a Third-Party AMD API<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">There are three common ways to get AI AMD, and each suits a different team:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Plug in an AMD API:<\/b><span style=\"font-weight: 400;\"> the fastest start. You pay per call, call audio leaves your network, and you depend on the vendor&#8217;s model.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Buy a dialer with AI AMD built in:<\/b><span style=\"font-weight: 400;\"> quick to launch, but you work within that product&#8217;s features, data model and pricing.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Build it into your own Asterisk dialer:<\/b><span style=\"font-weight: 400;\"> the most control over models, data, routing and cost at scale. It needs engineering effort and ongoing tuning.<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">If you are weighing a full build, our guide on how to<\/span> <a href=\"https:\/\/www.capanicus.com\/blog\/predictive-dialer-software-development\/\"><span style=\"font-weight: 400;\">build a predictive dialer<\/span><\/a><span style=\"font-weight: 400;\"> covers the architecture and features in more detail.<\/span><\/p>\n<h3><b>How Capanicus Adds AI AMD to Asterisk Dialers<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Capanicus builds custom predictive dialers and telecom platforms on Asterisk and FreeSWITCH. We can add AI answering machine detection to an existing Asterisk dialer or build it into a new one.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">That covers the audio streaming layer, the detection service, fallback to AMD(), routing rules per campaign, and reporting on false positives, NOTSURE share and decision time. Our<\/span> <a href=\"https:\/\/www.capanicus.com\/asterisk-freeswitch-development\"><span style=\"font-weight: 400;\">Asterisk development services<\/span><\/a><span style=\"font-weight: 400;\"> team works with your current dialplan, carriers and CRM, so you keep what already works.<\/span><\/p>\n<h2><b>Final Thoughts on AI AMD for Asterisk Dialers<\/b><\/h2>\n<p><span style=\"font-weight: 400;\">Answering machine detection decides whether your agents spend the day talking to people or to voicemail. Asterisk&#8217;s AMD() is a solid free starting point, but its timing rules struggle with short greetings, formal greetings, mobile voicemail and noisy lines.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">AI answering machine detection reads the audio itself, which helps exactly where timing rules fail. It works best when you test it on your own calls, measure false positives first and keep decision time short. Start in shadow mode, compare the results with AMD(), and roll it out one campaign at a time.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Planning AI AMD for your dialer?<\/span> <a href=\"https:\/\/www.capanicus.com\/contact-us\"><span style=\"font-weight: 400;\">Talk to our dialer engineers<\/span><\/a><span style=\"font-weight: 400;\"> about your Asterisk setup.<\/span><\/p>\n<h2><b>Frequently Asked Questions About AI Answering Machine Detection<\/b><\/h2>\n<h3><b>Does Asterisk have built-in answering machine detection?<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Yes. The AMD() dialplan application analyzes silence and word timing after a call is answered. It then sets <code style=\"background:#f3f4f6;color:#b3261e;padding:2px 5px;border-radius:4px;font-family:Consolas,Monaco,monospace;font-size:0.92em;\">AMDSTATUS<\/code> to HUMAN, MACHINE, NOTSURE or HANGUP.<\/span><\/p>\n<h3><b>Is AI AMD more accurate than Asterisk AMD()?<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">It often is on short, formal or unusual greetings, but results depend on the training data. Test both on your own recorded calls before you switch.<\/span><\/p>\n<h3><b>How fast should answering machine detection decide?<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">As fast as it can stay accurate. US rules count a call as abandoned if a live person is not connected within two seconds of their completed greeting.<\/span><\/p>\n<h3><b>Can I add AI AMD to my existing Asterisk dialer?<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">Usually, yes. The detection service connects at the Asterisk layer through AudioSocket, ARI External Media or EAGI, so the dialer and agent screens can stay the same.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Every outbound team knows the cost of a wrong answering machine call. The dialer connects an agent to a voicemail greeting, or worse, hangs up on a real person who just said &#8220;Hello?&#8221;. On Asterisk, the built-in AMD() application has handled this job for years, but its timing rules struggle with today&#8217;s voicemail greetings, mobile<\/p>\n","protected":false},"author":1,"featured_media":2645,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[175,173,174,140,177],"class_list":["post-2627","post","type-post","status-publish","format-standard","has-post-thumbnail","category-uncategorized","tag-ai-answering-machine-detection","tag-answering-machine-detection","tag-asterisk-amd","tag-predictive-dialer","tag-voicemail-detection"],"_links":{"self":[{"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/posts\/2627","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/comments?post=2627"}],"version-history":[{"count":2,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/posts\/2627\/revisions"}],"predecessor-version":[{"id":2647,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/posts\/2627\/revisions\/2647"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/media\/2645"}],"wp:attachment":[{"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/media?parent=2627"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/categories?post=2627"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.capanicus.com\/blog\/wp-json\/wp\/v2\/tags?post=2627"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}