Streaming synthesis
Start delivering audio after the first synthesized sentence instead of waiting for the entire answer to finish.
Local speech for Hermes Agent
Give Hermes local STT and TTS.
Install the published Python provider and let Hermes reuse AverVOX for streaming speech synthesis, transcription, capability discovery, and cancellation—without a separate cloud speech account.
Published on PyPITTS + STTStreamingWarm-daemon fallback
What it gives you
Use Hermes’s native speech interface with a local engine.
The provider translates Hermes requests into the AverVOX bridge contract, while preserving host-native configuration and lifecycle behavior.
Local transcription
Hand voice recordings to the same local Faster Whisper pipeline used by the AverVOX desktop workflow.
Capability-aware behavior
Detect the installed edition, engines, voices, STT model, daemon status, and feature support at runtime.
Setup
Use the self-checking path first.
The integration installer is designed to catch the most common false-positive setup: configuration exists, but the avrvx command or voice model is unavailable.
Verify AverVOX locally
Run a normal synthesis command and confirm the expected voice works on the machine where Hermes will call it.
Install the Hermes provider
Use avrvx --install-integration hermes for a guided setup, or install avervox-hermes directly from PyPI when managing configuration yourself.
Start the warm daemon when needed
For agents that speak frequently, keep the speech model loaded behind the private socket and let the provider fall back to one-shot processes if it stops.
Reference benchmark
Sub-second first audio with the warm bridge on the documented test machine.
The benchmark runs through Hermes’s real provider interface rather than timing an isolated Piper command.
Warm0.32sFirst audio
Warm0.77sWhole reply
Spawn2.6sFirst audio
Spawn3.3sWhole reply
Common questions
What to know before you start.
Which versions were tested?
The published provider documentation records Hermes 0.19.0, AverVOX 0.5.0 or newer, and Python 3.12. Verify the current package README when upgrading.
Does Hermes need to know whether OSS or Pro is installed?
No separate host package is required. The provider calls avrvx --capabilities and adapts to the engines and voices the installed edition exposes.
What file formats can be transcribed?
The bridge accepts recordings decodable by the installed ffmpeg path and prints the transcript on stdout. A no-speech result exits non-zero so the host can distinguish silence from success.
What happens on cancellation?
The bridge handles termination cleanly and returns a distinct cancellation exit status, allowing the host to stop playback or synthesis without reporting it as an ordinary failure.
Is a cloud speech API key required?
No. TTS and STT are executed by the local AverVOX installation. Hermes may still use whatever model or remote services you separately configure for its agent work.
Start with the workflow
Use one local speech engine across Hermes and the desktop.
Install the published provider, verify a real audio round-trip, and enable the daemon when repeated speech latency matters.