Answering Machine Detection AMD

Answering Machine Detection

LumenVox's new Answering Machine Detection (AMD) technology delivers outbound messages by detecting answering machine and voicemail tones. The AMD is tested at over 98%* accuracy.

Our tone-based detection is a significant innovation compared to many other energy-based answering machine detection technologies. The Lumenvox AMD actually listens to tones facilitating more reliable and highly accurate message delivery. Since LumenVox is listening for the beep, your outbound dialing application won't know it's connected to a voicemail system at the beginning of a call. With higher tone-detection accuracy your message sounds more professional because it starts precisely where it should.

Other answering machine detection systems attempt to measure the duration of the called party's speech to determine whether it is a live person or a machine. This is a highly imprecise method prone to failure. The new LumenVox AMD is simple and effective.

Available in the LumenVox Speech Engine, the new Answering Machine Detection is an effective method of verifying that an outbound call has been connected to a live person or an automated recording system.

The technology is compatible with most voice platforms and PBX systems, and fully supported through the standards-based Media Resource Control Protocol (MRCP). Since MRCP is supported by almost every major voice platform, it makes plugging answering machine detection into your application simple.

Specific Answering Machine Detection licenses are required to implement the tone detection.

Contact LumenVox today for information on purchasing answering machine detection licenses.

Using LumenVox Answering Machine Detection

To enable answering machine detection, you simply load a special grammar file that tells the Speech Engine to listen for beeps in the audio. As soon as it detects a tone, the Engine returns the event to your application or platform.

By loading another special grammar file, you can switch the Speech Engine back to speech recognition mode, letting you toggle between listening for beeps and listening for speech within a single call.

The guiding principle behind the technology is that your outbound application can treat both humans and answering machines the same up to the moment the beep occurs. To implement this technology, perform the following steps:

  1. Initiate your outbound call.
  2. When the called party side establishes a connection, wait a second or two and then play your message. (If you are connected to a voicemail system, your application may be playing audio at the same time the answering machine greeting is playing. This is acceptable since the machine at the other end will not be offended if your application is talking over it.)
  3. If a beep is detected, immediately stop playing your message. Then restart your message — or switch to a custom message intended for voicemail systems.
  4. When the voicemail owner listens to their messages, they will only hear the message you left after the beep.

This is a simple and effective method of leaving messages. Humans who answer hear the message as normal, and voicemail systems also receive an appropriate message.

Technical Specifications

LumenVox's answering machine detection technology runs on any system supported by the LumenVox Speech Engine, version 9.5 or newer. This means it can run on any modern Windows release or supported Linux distribution.

Answering machine detection is supported over MRCP, both versions 1 and 2. It should be compliant with any VoiceXML platform that also supports MRCP.

We have designed the technology to be easily dropped into existing platforms. From a technical perspective, you are just loading a special grammar file and waiting for a return from the Speech Engine.

System resource requirements are relatively minimal, but may scale up for very large numbers of simultaneous detections. This extra system load is in addition to any speech recognitions the Speech Engine may be doing.

*98% measured in large scale testing, based on recordings of actual calls. Accuracy may vary.

© 2016 LumenVox, LLC. All rights reserved.