***

title: Voice & Speech
sidebarTitle: Voice & Speech
icon: microphone
description: 'Industry-leading voice AI with Lightning, Pulse, and Electron'
----------------------------------------------------------------------------

Atoms is powered by our proprietary voice AI stack — the fastest, most accurate, and most natural-sounding in the industry. Three models working together in sequence: transcribe speech, reason through the response, synthesize voice. Sub-800ms end-to-end.

***

## Pulse — Speech-to-Text

High-accuracy, low-latency ASR built for real-time transcription. 32 languages with automatic detection.

| Spec               | Performance                    |
| ------------------ | ------------------------------ |
| **Latency (TTFT)** | 64ms                           |
| **English WER**    | 4.5%                           |
| **Best WER**       | 3.0% (Italian), 3.2% (Spanish) |
| **Languages**      | 32 supported                   |
| **Concurrency**    | 100 requests per GPU           |

**Key strengths:**

* 64ms time-to-first-transcript
* Industry-leading accuracy for Romance and Indic languages
* PII/PCI redaction built-in
* Handles accents and noisy environments

### Supported Languages

English, Hindi, Spanish, Portuguese, Italian, French, German, Dutch, Russian, Ukrainian, Polish, Czech, Slovak, Romanian, Bulgarian, Hungarian, Finnish, Swedish, Danish, Lithuanian, Latvian, Estonian, Maltese, Kannada, Malayalam, Telugu, Tamil, Marathi, Gujarati, Bengali, Punjabi, Oriya

***

## Electron — Small Language Model

Our optimized SLM for voice AI. Fast reasoning with low latency, purpose-built for conversational agents.

| Spec                 | Performance              |
| -------------------- | ------------------------ |
| **Optimized for**    | Voice conversations      |
| **Latency**          | Sub-500ms responses      |
| **Context handling** | Multi-turn conversations |

Electron understands conversational context, handles interruptions gracefully, and generates responses optimized for spoken delivery — not just text.

***

## Lightning v3.1 — Text-to-Speech

The fastest high-fidelity TTS model available. 44kHz native resolution with ultra-low latency.

| Spec              | Performance                               |
| ----------------- | ----------------------------------------- |
| **Latency**       | 175ms @ 20 concurrency                    |
| **Sample Rate**   | 44,100 Hz native                          |
| **Speed Control** | 0.5x to 2.0x                              |
| **Languages**     | English, Hindi (more coming)              |
| **Voice Cloning** | Instant (5-15s) and Professional (45min+) |

**Key strengths:**

* Studio-grade 44kHz audio clarity
* Natural prosody and intonation
* Real-time streaming (HTTP, SSE, WebSocket)
* Voice cloning for custom brand voices

***

## Get Started

<CardGroup cols={2}>
  <Card title="Platform" icon="browser" href="/atoms/atoms-platform/single-prompt-agents/agent-settings/voice-settings">
    Configure voice settings visually
  </Card>

  <Card title="Developer Guide" icon="code" href="/atoms/introduction/capabilities/voice-and-speech">
    Programmatic voice configuration
  </Card>
</CardGroup>