Public Transcriber Options
| Display name in docs | Selector label | Provider | Model id | Best for |
|---|---|---|---|---|
| Deepgram Nova 3 | Deepgram Nova 3 | Deepgram | nova-3 | Default production speech to text, English calls, and broad testing. |
| Deepgram Nova 2 | Deepgram Nova 2 | Deepgram | nova-2 | Existing agents tuned on Nova 2 or regression checks. |
| Deepgram Flux General EN | Deepgram Flux (English only) | Deepgram | flux-general-en | English-only calls that need sharper turn boundaries. |
| Soniox STT RT v4 | Soniox | Soniox | stt-rt-v4 | Hindi-English and code-switched conversations. |
Selection Rules
| Rule | Effect |
|---|---|
| New agents default to Nova 3 where no saved transcriber exists. | Users start with a known listening path. |
| Flux is English only in the dashboard. | Non-English language choices are not a supported match for Flux. |
| Soniox exposes Response Eagerness. | The Speech Settings panel shows the patient-to-eager timing slider for supported Soniox paths. |
| Fallback STT must use a different selectable transcriber. | The current selector excludes AssemblyAI from primary and fallback choices, and the fallback selector also excludes the current primary transcriber. |
| Transcriber pricing can appear in the selector. | Workspaces using rate preview can see ₹x.xx/min beside options. |
| Call logs snapshot the model used. | Billing and debugging can refer to the exact transcriber model on the call. |
Transcriber Fields
| Field | Meaning |
|---|---|
id | DialNexa transcriber id used by agent versions and billing lookup. |
provider | Speech to text provider, publicly documented as Deepgram or Soniox. |
model_id | Provider model identifier such as nova-3, nova-2, flux-general-en, or stt-rt-v4. |
name | Human-readable name shown by the product, with dashboard display labels applied where helpful. |
description | Provider model description. |
is_deleted | Whether the transcriber is unavailable for new selection. |
Which Transcriber Should I Start With?
| Question | Start with |
|---|---|
| Is the caller mostly English? | Deepgram Nova 3. |
| Is turn-taking the main English problem? | Deepgram Flux for English, then compare against Nova 3. |
| Will callers mix Hindi and English? | Soniox with Hindi-English language setup. |
| Are post-call fields wrong? | Review transcript quality before changing extraction fields. |
| Are names or product words wrong? | Test provider, language, recording quality, and prompt examples. |
| Is primary STT sometimes slow? | Enable fallback STT with a different provider and start with a 500 ms fallback wait. |
Related Reading
Speech To Text
Learn how transcribers affect live and post-call behavior.
Speech Settings
Tune Response Eagerness and Denoising Mode.
Latency And Turn Taking
Diagnose timing and interruption issues.
Provider Selection Guide
Choose the complete Voice AI stack.