OmniVoice

Batteries-included voice pipeline framework for Go. This package provides a unified interface for speech-to-text (STT) and text-to-speech (TTS) with all providers included.

For a minimal dependency footprint, use omnivoice-core instead.

Voice Architecture

OmniVoice supports two approaches for real-time voice:

Approach	Latency	Use Case
Traditional Pipeline (STT→LLM→TTS)	500-1500ms	Custom voices, domain-specific STT
Native Voice-to-Voice	100-200ms	Low latency, natural conversation

This package provides the Traditional Pipeline components. For native voice-to-voice, see:

• omni-openai/omnivoice/realtime - OpenAI Realtime API (~100ms)
• omni-google/omnivoice/realtime - Gemini Live API (~200ms)
• omni-deepgram/omnivoice/realtime - Deepgram Voice Agent (~100-300ms)

See the Voice Architecture Guide for detailed comparison.

Features

• 🎯 Unified Interface: Single API for all STT and TTS providers
• 📦 Provider Registry: Get providers by name - no need to import individual provider packages
• 🔌 Multiple Providers: OpenAI, Deepgram, ElevenLabs, Twilio, Telnyx + local providers
• 💻 Local Voice: F5-TTS and Whisper via MLX on Apple Silicon (zero API cost)
• ⚡ Streaming Support: Real-time transcription and synthesis
• 🚀 Easy Integration: Import and use with minimal configuration

Installation

go get github.com/plexusone/omnivoice

CLI

OmniVoice includes a command-line tool for transcription.

Install CLI

go install github.com/plexusone/omnivoice/cmd/omnivoice@latest

Usage

# Set your API keyexport DEEPGRAM_API_KEY="your-api-key"# Basic transcription (stdout)
omnivoice transcribe podcast.mp3
# Save to file
omnivoice transcribe -p deepgram -o transcript.txt podcast.mp3
# JSON output with full metadata (OmniVoice Transcript format)
omnivoice transcribe -p deepgram --diarize --timestamps -f json -o transcript.json podcast.mp3
# Generate SRT subtitles
omnivoice transcribe -p deepgram -f srt -o subtitles.srt podcast.mp3
# Generate WebVTT subtitles
omnivoice transcribe -p deepgram -f vtt -o subtitles.vtt podcast.mp3
# List available providers
omnivoice providers list

Output Formats

Format	Description
text	Plain transcript text (default)
json	OmniVoice Transcript format with full metadata
srt	SubRip subtitles
vtt	WebVTT subtitles

Environment Variables

Variable	Provider
DEEPGRAM_API_KEY	Deepgram
OPENAI_API_KEY	OpenAI
ELEVENLABS_API_KEY	ElevenLabs

Quick Start (Library)

import (
"github.com/plexusone/omnivoice"
_ "github.com/plexusone/omnivoice/providers/all"// Register all providers
)

Usage

package main
import (
"context""log""os""github.com/plexusone/omnivoice"
_ "github.com/plexusone/omnivoice/providers/all"
)
funcmain() {
ctx:=context.Background()
// Get providers by name using the registrysttProvider, err:=omnivoice.GetSTTProvider("deepgram",
omnivoice.WithAPIKey(os.Getenv("DEEPGRAM_API_KEY")))
iferr!=nil {
log.Fatal(err)
}
ttsProvider, err:=omnivoice.GetTTSProvider("elevenlabs",
omnivoice.WithAPIKey(os.Getenv("ELEVENLABS_API_KEY")))
iferr!=nil {
log.Fatal(err)
}
// Transcribe audioresult, err:=sttProvider.TranscribeFile(ctx, "audio.mp3", omnivoice.TranscriptionConfig{
Language: "en",
EnableWordTimestamps: true,
})
iferr!=nil {
log.Fatal(err)
}
log.Printf("Transcription: %s", result.Text)
// Synthesize speechaudio, err:=ttsProvider.Synthesize(ctx, "Hello, world!", omnivoice.SynthesisConfig{
VoiceID: "pNInz6obpgDQGcFmaJgB", // Adam
})
iferr!=nil {
log.Fatal(err)
}
// audio.Audio contains the audio bytes
}

Provider Registry

Get providers by name at runtime - no need to import individual provider packages:

// STT/TTS providers: "openai", "elevenlabs", "deepgram", "twilio"ttsProvider, _:=omnivoice.GetTTSProvider("elevenlabs", omnivoice.WithAPIKey(key))
sttProvider, _:=omnivoice.GetSTTProvider("deepgram", omnivoice.WithAPIKey(key))
// Realtime providers: "openai", "gemini", "deepgram"rtProvider, _:=omnivoice.GetRealtimeProvider("openai", omnivoice.WithAPIKey(key))
// Gateway providers: "twilio", "telnyx"gateway, _:=omnivoice.GetGatewayProvider("twilio", omnivoice.WithAccountSID(sid), omnivoice.WithAuthToken(token))
// List registered providersfmt.Println(omnivoice.ListTTSProviders()) // [openai elevenlabs deepgram twilio]fmt.Println(omnivoice.ListSTTProviders()) // [openai elevenlabs deepgram twilio]fmt.Println(omnivoice.ListRealtimeProviders()) // [openai gemini deepgram]fmt.Println(omnivoice.ListGatewayProviders()) // [twilio telnyx]// Realtime factory API (for gateway integration)factory, _:=omnivoice.GetRealtimeFactory("openai")
fmt.Println(omnivoice.ListRealtimeFactories()) // [openai gemini deepgram]

Language Codes

OmniVoice accepts language codes in BCP-47 format, which includes ISO 639-1 two-letter codes and regional variants.

Common codes:

Code	Language
en	English
en-US	English (US)
en-GB	English (UK)
es	Spanish
es-MX	Spanish (Mexico)
fr	French
de	German
it	Italian
pt	Portuguese
pt-BR	Portuguese (Brazil)
ja	Japanese
ko	Korean
zh	Chinese
zh-CN	Chinese (Simplified)
zh-TW	Chinese (Traditional)
ar	Arabic
hi	Hindi
ru	Russian

Notes:

• Use simple codes (en) for broad compatibility across providers
• Use regional variants (en-US) when accent/dialect matters for TTS
• Provider support varies; see provider documentation for full language lists
• STT providers generally support automatic language detection when no code is specified

Included Providers

Cloud STT/TTS Providers

Provider	STT	TTS	Registry Name
OpenAI	Whisper	TTS-1/TTS-1-HD	"openai"
ElevenLabs	Scribe	Multilingual v2	"elevenlabs"
Deepgram	Nova-2	Aura	"deepgram"
Twilio	Media Streams	Media Streams	"twilio"

Local Providers (Apple Silicon)

Provider	Type	Registry Name	Requirements
F5-TTS MLX	TTS	"f5tts-mlx"	Apple Silicon, Python server
Whisper MLX	STT	"whisper-mlx"	Apple Silicon, Python server

Local providers are opt-in imports from omnivoice-core. See Local TTS Guide.

Native Voice-to-Voice (Realtime)

Provider	Latency	Registry Name
OpenAI Realtime	~100ms	"openai"
Gemini Live	~200ms	"gemini"
Deepgram Voice Agent	~100-300ms	"deepgram"

Voice Gateway

Provider	Registry Name
Twilio	"twilio"
Telnyx	"telnyx"

Related Packages

Core

• omnivoice-core - Core interfaces (minimal dependencies)

Local Providers (Apple Silicon)

Available in omnivoice-core v0.15.0+ for on-device voice processing:

• f5tts-mlx - F5-TTS via MLX (zero API cost)
• whisper-mlx - Whisper via MLX (zero API cost)

Native Voice-to-Voice (Recommended for Low Latency)

• omni-openai/omnivoice/realtime - OpenAI Realtime API (~100ms latency)
• omni-google/omnivoice/realtime - Gemini Live API (~200ms latency)
• omni-deepgram/omnivoice/realtime - Deepgram Voice Agent (~100-300ms latency)

STT/TTS Providers

• omni-openai - OpenAI provider (Whisper, TTS-1)
• omni-deepgram - Deepgram provider (Nova-2, Aura)
• elevenlabs-go - ElevenLabs SDK

Voice Gateway Providers

• omni-twilio - Twilio Media Streams
• omni-telnyx - Telnyx Media Streaming
• omni-vonage - Vonage Voice WebSocket
• omni-plivo - Plivo Stream API
• omni-livekit - LiveKit WebRTC (web/mobile)

License

MIT License - see LICENSE for details.