Release Notes: v0.9.0¶

Release Date: 2026-06-15

Highlights¶

RealtimeProvider Registry: Native voice-to-voice APIs with ~100-300ms latency
OpenAI Realtime & Gemini Live: Two providers registered out of the box

New Features¶

Realtime Provider Registry¶

Get native voice-to-voice providers by name at runtime:

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

// OpenAI Realtime API (~100ms latency)
rt, err := omnivoice.GetRealtimeProvider("openai-realtime",
    omnivoice.WithAPIKey(os.Getenv("OPENAI_API_KEY")))

// Google Gemini Live API (~200ms latency)
rt, err := omnivoice.GetRealtimeProvider("gemini-live",
    omnivoice.WithAPIKey(os.Getenv("GOOGLE_API_KEY")))

// List registered providers
fmt.Println(omnivoice.ListRealtimeProviders()) // [openai-realtime gemini-live]

Type Aliases¶

New types re-exported from omnivoice-core:

Type	Description
`RealtimeProvider`	Interface for voice-to-voice providers
`ProcessConfig`	Configuration for realtime sessions
`FunctionDeclaration`	Function the model can call
`RealtimeAudioChunk`	Audio data from the model
`RealtimeTranscript`	Transcript update during conversation
`RealtimeClient`	Multi-provider client with failover

Registry Functions¶

Function	Description
`RegisterRealtimeProvider`	Register a provider factory
`GetRealtimeProvider`	Create provider by name
`ListRealtimeProviders`	List registered providers
`HasRealtimeProvider`	Check if provider exists
`NewRealtimeClient`	Create multi-provider client

Voice Architecture Comparison¶

Approach	Latency	Use Case
Native Voice-to-Voice	100-300ms	Low latency, natural conversation
Traditional Pipeline (STT->LLM->TTS)	500-1500ms	Custom voices, domain-specific STT

When to Use Native Voice-to-Voice¶

Real-time voice agents requiring <500ms response time
Natural conversational experiences
Applications where latency is critical

When to Use Traditional Pipeline¶

Custom voice synthesis (ElevenLabs voices)
Domain-specific STT models (medical, legal)
Multi-provider flexibility

Dependencies¶

Package	Change
`omni-twilio`	v0.5.0 -> v0.6.0 (migrated from twilio-go)
`omni-openai`	v0.2.2 -> v0.3.0 (Realtime API support)
`omni-google`	Added v0.5.0 (Gemini Live API)
`omnivoice-core`	v0.9.0 -> v0.13.0 (Realtime package)

Installation¶

go get github.com/plexusone/omnivoice@v0.9.0

Migration Guide¶

From v0.8.0¶

No breaking changes. New features are additive:

Realtime providers: Import from github.com/plexusone/omnivoice
Provider registration: Use _ "github.com/plexusone/omnivoice/providers/all" (includes new providers)

Quick Start¶

package main

import (
    "context"
    "os"

    "github.com/plexusone/omnivoice"
    _ "github.com/plexusone/omnivoice/providers/all"
)

func main() {
    ctx := context.Background()

    // Get OpenAI Realtime provider
    rt, err := omnivoice.GetRealtimeProvider("openai-realtime",
        omnivoice.WithAPIKey(os.Getenv("OPENAI_API_KEY")))
    if err != nil {
        panic(err)
    }

    // Start a realtime session
    session, err := rt.Connect(ctx, omnivoice.ProcessConfig{
        Instructions: "You are a helpful voice assistant.",
    })
    if err != nil {
        panic(err)
    }
    defer session.Close()

    // Process audio...
}

Full Changelog¶

See CHANGELOG.md for the complete list of changes.