Live Speaker Detection
Live Speaker Detection allows you to classify whether the number you're calling is answered by a human, answering machine or fax machine. That's useful if, for example, you want to ask someone a question or, if they're out, leave them a message.
Live Speaker Detection is commonly referred to as Answering Machine Detection (AMD). In the REST API, we call it Far-end Classification. Whichever term is used, they're one and the same thing.
The Live Speaker Detection timeout is the longest you're prepared to wait for it to classify. The maximum value is 5000 msec. That gives the best results and, unless the country you're dialling has a shorter legal restriction, it's generally the best value to use.
There are three Live Speaker Detection modes:
The REST API always uses Default mode - its Live Speaker Detection is not configurable.
The UAS API uses Default mode unless you specify a different one and, in most cases, Default is the best mode to use.
If you're dialling through Aculab's outbound PSTN provider, Default mode chooses the best mode based on the country code of the number you dial. Otherwise, it assumes the best mode based on the Cloud you're using - Network Connect for the Europe Cloud and Answer Detect for USA Clouds.
Network Connect and Answer Detect modes
These modes are explained in the tabs below.
Live Speaker Detection starts when the call is connected.
This mode is best suited to dialling numbers with non-US country codes. The telephone systems in these countries connect the call when the callee or their answering service picks up the phone
Answer Detection starts when the call is connected, then Live Speaker Detection starts when the callee picks up the phone.
This mode is best suited to dialling numbers with US country codes. Some US telephone systems connect the call as soon as they start the callee's line ringing. This mode holds off Live Speaker Detection until the callee's line is picked up.
Our Live Speaker Detection has been extensively optimised using a large set of data and performs well on high volumes of calls every day. However, please note that, due to the huge variability in audio content on answered calls, this task inherently cannot provide complete accuracy.