Agent Voice Response vs JVoiceXML

Agent Voice Response is a self-hosted system that turns an Asterisk PBX into a real-time voice AI agent. Callers can interrupt while it speaks through voice-activity detection. Its composable microservices let operators swap automatic speech recognition, language-model, and text-to-speech engines. It supports Asterisk AudioSocket on Asterisk 18 and later, including named systems such as FreePBX, VitalPBX, Vicidial, Elastix, and custom dialplans. Listed integrations include OpenAI, Anthropic, Deepgram, ElevenLabs, Google, Ollama, Vosk, Kokoro, Ultravox, OpenRouter, CoquiTTS, and Silero. Noise cancellation and echo suppression are described before audio reaches the model. Local deployments can use Vosk or Silero, Ollama, and CoquiTTS or Kokoro, keeping caller audio on the network according to the maker. Docker microservices can scale horizontally behind a load balancer. AVR is free for personal and commercial use, with no license keys or per-minute charge. Setup requires Docker images, provider configuration, and an Asterisk dialplan extension.

JVoiceXML is a free, open-source VoiceXML interpreter for Java, described by its project as implementing VoiceXML 2.1. It supports dialogs that combine synthesized speech or digitized audio with spoken or DTMF input. Dialog functions also include recording, telephony, and mixed-initiative conversations. The architecture allows custom extensions, and demo implementation platforms support Java APIs such as JSAPI and JTAPI. The repository includes modules for Mary, MRCPv2, JSAPI 2.0, JTAPI, and text implementations. The project names platform independence as a major goal. Its repository identifies the license as LGPL-2.1 and provides release downloads. The latest listed version is 0.7.8, which added an SRGS grammar processor with SISR support, Microsoft LUIS as a grammar processor, bug fixes, and a move to Gradle. JVoiceXML is listed for API, Linux, and self-hosted use.