Logo
Search
API Docs

Voice Fallback Plans

Voice & Audio Tuning

Voice Fallback Plans: Multi-Provider TTS Failover

Overview

A voice fallback plan lets you configure backup voices that automatically activate if your primary voice provider fails mid-call. Instead of the call dropping or going silent, Sulus switches to the next voice in your fallback list, so the conversation continues without the caller ever noticing an outage.

You configure this via the fallbackPlan property on your assistant's voice object, listing fallback voices in priority order within the voices array.


How Fallback Works

When a voice failure occurs during a call, Sulus follows this sequence:

  1. Detects the failure of the primary voice provider
  2. Switches to the first fallback voice in your voices array
  3. Continues sequentially through the list if subsequent failures occur
  4. Terminates the call only if all voices in the plan have failed

Configuring a Fallback Plan

Add the fallbackPlan property to your assistant's voice object and list fallback voices in the voices array:

{
  "voice": {
    "provider": "openai",
    "voiceId": "shimmer",
    "fallbackPlan": {
      "voices": [
        {
          "provider": "cartesia",
          "voiceId": "248be419-c632-4f23-adf1-5324ed7dbf1d"
        },
        {
          "provider": "11labs",
          "voiceId": "cgSgspJ2msm6clMCkdW9",
          "stability": 0.5,
          "similarityBoost": 0.75
        }
      ]
    }
  }
}

ElevenLabs Settings

When using 11labs as a fallback voice, the following settings are available:

SettingDescription
stabilityControls consistency of the voice's tone across a sentence
similarityBoostHow closely the output matches the original voice sample
styleAmount of stylistic exaggeration applied to the voice
useSpeakerBoostBoosts similarity to the target speaker at a small latency cost
speedPlayback speed of the synthesized speech
optimizeStreamingLatencyTrades audio quality for lower streaming latency
enableSsmlParsingEnables SSML tags in the text sent to this voice
modelWhich ElevenLabs model to use for synthesis

Cartesia Settings

When using Cartesia as a fallback voice, the following settings are available:

SettingDescription
modelWhich Cartesia model to use
languageTarget language for synthesis
experimentalControls.speedAdjusts playback speed
experimentalControls.emotionAdjusts emotional tone of the voice
generationConfigAdditional generation parameters, used specifically with the sonic-3 model

Azure, OpenAI, LMNT, and PlayHT Settings

ProviderAvailable Settings
Azurespeed
OpenAIspeed, model, instructions
LMNTspeed, language
PlayHTspeed, temperature, emotion, voiceGuidance, styleGuidance, textGuidance, model

Deepgram, Rime AI, Hume, Minimax, WellSaid, Neuphonic, and SmallestAI Settings

ProviderAvailable Settings
Deepgrammodel, mipOptOut
Rime AImodel, speed, pauseBetweenBrackets, phonemizeBetweenBrackets, reduceLatency, inlineSpeedAlpha
Humemodel, description, isCustomHumeVoice
Minimaxmodel, emotion, pitch, speed, volume
WellSaidmodel, enableSsml, libraryIds
Neuphonicmodel, language, speed
SmallestAImodel, speed

Best Practices

  • Use different providers across your fallback voices to protect against provider-wide outages
  • Select voices with similar characteristics (tone, accent, gender) to maintain a consistent caller experience
  • Sulus recommends configuring 2–3 fallbacks from different providers for optimal reliability

In summary, a fallback plan is a safety net for the voice layer of your assistant — it keeps calls running smoothly even if a primary voice provider has an outage, by automatically stepping through a prioritized list of backup voices.