Does voice mode work offline?

Local STT (faster-whisper) works offline. Cloud providers need internet. Edge TTS needs internet but is free.

Can I use voice in Discord/Slack?

Voice message transcription works on all platforms. TTS output as audio messages depends on platform support.

How much VRAM does local STT need?

The 'base' model needs ~1GB. 'large-v3' needs ~10GB. Most laptops can run 'base' or 'small' comfortably.

Voice Mode Setup — Speak to Hermes, Hear Responses

Voice mode transforms Hermes into a spoken assistant. Speak your commands instead of typing, hear responses read aloud, and have voice messages automatically transcribed on messaging platforms.

Components

Speech-to-Text (STT): Your voice → Text commands Text-to-Speech (TTS): Hermes responses → Audio

STT Providers

Provider	Speed	Cost	Privacy
Local (faster-whisper)	Good	Free	Full
Groq	Excellent	Free tier	Cloud
OpenAI Whisper	Good	Paid	Cloud
Mistral Voxtral	Good	Paid	Cloud

TTS Providers

Provider	Quality	Cost
Edge TTS	Good	Free
ElevenLabs	Excellent	Paid
OpenAI TTS	Excellent	Paid
MiniMax	Good	Paid
Mistral	Good	Paid

Quick Setup: Free Local Voice

1. Install voice dependencies:

pip install 'hermes-agent[voice]'

2. Enable in config:

stt:
  enabled: true
  provider: local
  local:
    model: base  # tiny | base | small | medium | large-v3

3. Use in CLI:

/voice on
# Press Ctrl+B to record
# Speak your command
# Release to send

Voice on Messaging Platforms

Voice messages are auto-transcribed:

[Telegram voice note] → "Hey Hermes, check the server status"

Hermes processes the text normally, responds with text.

To enable TTS responses on messaging:

tts:
  enabled: true
  provider: edge  # Free Microsoft voices

Model Size vs Quality

Model	Size	VRAM	Quality
tiny	39M	~1GB	Fast, less accurate
base	74M	~1GB	Balanced
small	244M	~2GB	Better accuracy
medium	769M	~5GB	Good quality
large-v3	1550M	~10GB	Best quality
turbo	809M	~6GB	Fast + good

Language Detection

Auto-detect language:

stt:
  local:
    model: base
    # language: ""  # Leave empty for auto-detect

Force specific language:

stt:
  local:
    model: base
    language: "en"  # or "es", "fr", "de", etc.

Cloud STT with Groq (Free)

# ~/.hermes/.env
GROQ_API_KEY=gsk_xxx

stt:
  enabled: true
  provider: groq  # Uses Whisper via Groq's free tier

Premium TTS with ElevenLabs

ELEVENLABS_API_KEY=el_xxx

tts:
  enabled: true
  provider: elevenlabs

Troubleshooting

"No audio input detected"

Check microphone permissions
Verify default input device
In CLI, ensure Ctrl+B triggers recording

"STT transcription inaccurate"

Use larger model: model: small or model: large-v3
Force language if known: language: "en"

"TTS sounds robotic"

Switch from Edge TTS to ElevenLabs or OpenAI
These have more natural voices

Voice Mode: Hands-Free AI Assistance