Public Voice Providers
| Provider | What users see | Best fit |
|---|---|---|
| ElevenLabs | Voice rows, sample playback, Nexa voice id, supported languages, and Flash v2.5 model selection where available. Default model: eleven_flash_v2_5. | Brand personality, broad voice auditioning, and pronunciation-sensitive use cases. |
| Cartesia | Cartesia voice rows, language-aware voices, model options, speed, stability, and volume controls. Model: sonic-2. | Fast streamed speech, clean call audio, and production agents that need a natural voice for a specific language. |
| SmallestAI | Indian voice personas (Diya, Raman, Ananya, Aarav, and more). Models: lightning, lightning-large, lightning-v2. | Indian language agents, Hindi, Hinglish, and Indian English calls. |
| Sarvam AI | India-focused TTS with Indian English and regional language support. Model: bulbul:v2, default language: en-IN. | Indian English callers and India-focused deployments. |
Voice Fields
| Field | Meaning |
|---|---|
provider | Public voice provider: ElevenLabs, Cartesia, SmallestAI, or Sarvam AI. |
provider_voice_id | Provider-specific voice identifier. |
id | DialNexa voice id. The UI can also show it as voice_... or vel_... depending on context. |
name | Display name shown in the voice selector. |
accent | Accent label used for filtering when available. |
age_group | Approximate age group label where available. |
gender | Gender label used for filtering. |
languages | Language records linked to the voice. |
recording | Sample audio URL used by the play button. |
Voice Model Fields
| Field | Meaning |
|---|---|
provider | Voice model provider. |
provider_voice_model_id | Provider model id, such as eleven_flash_v2_5 for ElevenLabs Flash v2.5. |
name | Model display name in the voice settings popover. |
description | Optional model explanation shown in tooltips. |
pricing_per_minute | Per-minute pricing metadata when available. |
is_deleted | Whether the model is unavailable for selection. |
Dashboard Selection Behavior
| Behavior | What it means |
|---|---|
| Voice modal starts without a language filter. | Users see the full provider catalog first, then narrow it by language when needed. |
| Voices can be filtered by provider, language, gender, accent, and search. | Large voice lists stay usable. |
| Each voice row can have its own language dropdown. | Users choose the language for the selected voice before applying it. |
The Nexa voice ID copies as vel_.... | Support, API-adjacent setup, and internal notes can refer to the same voice. |
| ElevenLabs current dashboard path shows Flash v2.5 where supported. | Test the visible model instead of expecting older models to appear for new selections. |
| Published versions can lock voice settings. | Edit a draft version when testing voice changes. |
Voice Selection Checklist
Filter by language first
A good English sample does not prove the voice works for Hindi-English or another language.
Run a real test call
Use your greeting, names, amounts, dates, and compliance lines through the actual route.
Related Reading
Text To Speech
Compare ElevenLabs and Cartesia behavior.
Languages Voices Models And Transcribers
Choose the complete conversation stack.
Supported Languages
Match voices to language requirements.
Testing Agents
Listen before publishing.