Get started

Advanced Voice

Full voice I/O with OpenAI Whisper and 40+ ElevenLabs voices.

Tier: Standard | Price: $5 one-time

Overview

Advanced Voice adds full voice input and output to QuoxCORE. Speech-to-text is handled by OpenAI Whisper, and text-to-speech runs through ElevenLabs with a library of 40+ voices. Each agent in your workspace can be assigned a distinct voice with individually tuned parameters.

The plugin covers the complete voice pipeline: push-to-talk audio capture, transcription, AI response summarisation for spoken delivery, and configurable TTS output. Premium tier users get an extended 2,000 character TTS limit per response.

Speech-to-Text

Voice input uses OpenAI's Whisper model for transcription.

Push-to-Talk

The VoiceButton component activates via three input methods:

  • Mouse -- click and hold to record
  • Touch -- press and hold on mobile or tablet
  • Spacebar -- hold the spacebar while focused on the voice button

Each recording can be up to 30 seconds long. Release the button to stop recording and send the audio for transcription.

Audio Capture Flow

  1. The VoiceButton triggers the browser's MediaRecorder API.
  2. Audio is captured in WebM format.
  3. On release, the recording is sent as a POST request to /webhook/transcribe.
  4. Whisper processes the audio and returns the transcribed text.
  5. The transcription is inserted into the chat input, ready for editing or sending.

No audio is stored after transcription. The WebM data is discarded once the Whisper response is received.

Text-to-Speech

Spoken responses use ElevenLabs with over 40 available voices across multiple categories.

Voice Categories

Voices are organised into groups by character and tone. Categories include general-purpose narration voices, conversational voices, and character voices suited to specific personas. Each voice has a preview available in the plugin settings panel.

TTS Models

Three ElevenLabs models are available, each with a different latency and quality profile:

ModelCharacteristicsBest for
FlashLowest latency, slightly reduced qualityReal-time interactions where speed matters
TurboBalanced latency and qualityGeneral use, the default
Multilingual v2Highest quality, higher latencyPolished output, multilingual content

Select the model per-agent or set a workspace default in plugin settings.

Per-Agent Voice Assignment

Up to 49 agents can each be assigned a distinct voice. This makes multi-agent conversations audible without confusion -- each agent sounds different.

To assign a voice:

  1. Open the Advanced Voice settings panel.
  2. Select an agent from the agent list.
  3. Choose a voice from the library and preview it.
  4. Adjust parameters (see below) for that agent.
  5. Save. The assignment takes effect immediately.

Agents without a voice assignment fall back to the workspace default voice.

Parameter Tuning

Each voice assignment exposes four adjustable parameters:

ParameterRangeEffect
Stability0.0 -- 1.0Higher values produce more consistent, predictable speech. Lower values introduce more variation and expressiveness.
Similarity Boost0.0 -- 1.0Controls how closely the output matches the original voice sample. Higher values sound more like the reference voice.
Style0.0 -- 1.0Adjusts the emotional expressiveness of the delivery. Higher values produce more animated speech.
Speed1.0x -- 1.5xPlayback speed multiplier. 1.0x is normal pace, 1.5x is the maximum.

Parameters are stored per agent. Experiment with stability and similarity boost first, as these have the most noticeable effect on output quality.

Verbosity Modes

The plugin offers three verbosity modes that control how much of the AI response is spoken aloud:

ModeCharacter LimitUse case
Quiet150 charactersBrief confirmations and short answers
Normal500 charactersStandard responses, the default
Verbose1,000 charactersDetailed explanations and longer output

Premium tier extends the maximum TTS limit to 2,000 characters regardless of mode.

Switch modes from the voice control bar or via plugin settings. The mode affects spoken output only -- the full text response always appears in the chat window.

Smart Summarisation

When an AI response exceeds the current verbosity limit, the plugin uses AI-powered summarisation to condense the text before sending it to TTS. This keeps spoken output concise and coherent rather than simply truncating mid-sentence.

Text Preprocessing

Before summarisation and TTS, the plugin strips content that does not translate well to speech:

  • Code blocks and inline code
  • Markdown tables
  • IP addresses
  • URLs and links

This preprocessing runs automatically. The original response text in the chat window is unaffected.

Configuration

API Keys

Advanced Voice requires two API keys:

  • OpenAI API key -- for Whisper transcription
  • ElevenLabs API key -- for TTS voice synthesis

Enter both keys in the plugin settings panel. Keys are stored in your workspace configuration and are not shared across workspaces.

Export and Import

Voice configurations, including agent-voice assignments, parameter tuning, and verbosity settings, can be exported as a JSON file. This is useful for:

  • Backing up your voice setup
  • Replicating configuration across QuoxCORE instances
  • Sharing voice presets with team members

To export: open plugin settings, click Export Configuration, and save the JSON file. To import: click Import Configuration and select a previously exported file. Existing settings are overwritten on import.