99.7% accurate answering machine detection powered by proprietary neural networks trained on millions of real voicemail interactions. SpeechLLM doesn't just measure timing — it understands what's being said.
Raw telephony audio captured from carrier network and normalized across volume levels and codec variations.
Audio converted to mel-spectrograms — 2D visual representations encoding frequency and time information.
Transformer-based neural network with self-attention mechanisms analyzes the spectrogram across multiple dimensions simultaneously.
Returns HUMAN, VOICEMAIL, CALLGUARD, IVR, or DISCONNECT with confidence score and response time <50ms.
WebSocket API works seamlessly with VICIdial, Asterisk, FreeSwitch, GoAutoDial, and any dialer supporting audio streaming:
Call Answered ↓ Audio streams to VM Hunter via WebSocket ↓ SpeechLLM analyzes in <50ms ↓ Decision returned: HUMAN / MACHINE / CALLGUARD ↓ AMDSTATUS + AMDCAUSE + CONFIDENCE set ↓ Your dialplan routes accordingly
No infrastructure replacement needed. No dialplan rewrites. Standard AMD variable compatibility.
Test SpeechLLM on your real calls. No credit card required. Deploy to production in minutes.
Start Your Free TrialWant to dive deeper into the technology?
Read: SpeechLLM's 8-Layer Architecture for AMD →
