Voice Profiles

Voice profiles connect ElevenLabs voices to the speech system. Each profile wraps an ElevenLabs voice with tuning parameters that control how the synthesized speech sounds. Voice profiles are referenced by Speech Pack templates to determine which voice delivers each announcement.

Backoffice path: backoffice/voice-profiles

Available to roles: System Administrator, Administrator

Creating a Voice Profile

When you open the create page, the system fetches the list of available voices from the ElevenLabs API. Each voice is presented with:

• Preview audio -- a short sample you can play to hear the voice
• Labels -- metadata tags such as accent (e.g., "British", "American"), age (e.g., "young", "middle-aged"), and style (e.g., "narrative", "conversational")

Select a voice to populate the profile. The name field is auto-filled from the ElevenLabs voice name, but you can change it.

Fields

Field	Description	Default / Range
Name	Display name for the profile	Auto-filled from ElevenLabs voice name
ElevenLabs Voice ID	The selected voice from the ElevenLabs API	Selected during creation
ElevenLabs Model ID	The TTS model to use	eleven_turbo_v2_5
Speech Language	Default Speech Language for digit pronunciation and salutations	Required
Stability	Controls voice consistency across generations. Lower values produce more expressive, varied output; higher values produce more predictable output	Discrete values: 0.0, 0.5, or 1.0 (for v3 model)
Similarity Boost	How closely the output matches the original voice sample. Higher values increase fidelity but may amplify artifacts	0.0 to 1.0
Style	Controls the expressiveness and emotional range of the voice	0.0 to 1.0
Speed	Playback speed multiplier for the generated audio	0.7 to 1.2
Use Speaker Boost	Toggle that enhances voice clarity and presence at the cost of slightly higher latency	On / Off
Active	Toggle to enable or disable this voice profile	On / Off

Fragment Count

The list view displays a fragment count for each voice profile. This shows how many cached audio files (speech fragments) currently exist that were generated using this voice. This is useful for understanding which voices are actively in use and how much cached content depends on each profile.

Tuning Guidelines

• Stability at 0.5 is a good starting point for most cabin announcement use cases -- it balances natural variation with consistency.
• Similarity Boost above 0.75 produces very close reproductions of the original voice but can introduce audio artifacts if the source voice sample is low quality.
• Style at 0.0 produces neutral, flat delivery. Values around 0.3 to 0.5 add natural inflection suitable for PA announcements without sounding overly dramatic.
• Speed at 1.0 is normal pace. Reduce to 0.85 - 0.9 for a calmer, more measured cabin announcement feel. Values below 0.7 or above 1.2 are not supported.
• Enable Speaker Boost for voices that sound too quiet or lack presence in the mix, especially when the PA microphone effect is applied in speech packs.

Usage Notes

• A voice profile must be active to be used by speech pack templates. Deactivating a profile does not delete cached fragments, but new audio will not be generated with that voice until it is reactivated.
• Each voice profile is linked to a single speech language by default. Individual speech pack templates can override this language if needed (e.g., a German voice profile used for an English-language announcement).
• Changing tuning parameters (stability, similarity boost, style, speed) does not automatically regenerate existing cached fragments. New fragments will use the updated settings, but previously cached audio remains unchanged until the underlying text changes.
• You cannot delete a voice profile that has cached fragments in use. Deactivate it instead if you want to retire a voice.