I am using Asterisk with a call to the amd() command.
I have added the "play a sound file before you call amd() and AMD() does detect audio and does determine human vs machine. The problem I am having is that as soon as amd detects a machine, it jumps to the machine context without waiting for the machine to “beep”, so it begins playing back the sound file before the beep and most of the message is cut off.
For debugging I NoOp TALK_DETECTED to see how long the answering machine talks, and in 1.4 this worked, now TALK_DETECTED is always NULL.

Here is my amd.conf file:
initial_silence = 3000
greeting = 4000
after_greeting_silence = 350
total_analysis_time = 5000
min_word_length = 120
between_words_silence = 50
maximum_number_of_words = 3
silence_threshold = 256

Anyone know how to correct the error and have amd populate TALK_DETECTED with a number that represents ms of audio like it did in 1.4?