> ## Documentation Index
> Fetch the complete documentation index at: https://docs.geekhub.mx/llms.txt
> Use this file to discover all available pages before exploring further.

# Text-to-Speech (TTS)

> Audio synthesis from text. /v1/audio/speech endpoint compatible with the OpenAI SDK. Real-time streaming, 6 voices, 6 formats.

Endpoint `/v1/audio/speech` is OpenAI-SDK compatible: same body shape, same response shape (binary audio with `Content-Type` per format). Supports full or streaming responses and respects org-level ZDR flags.

## Syntax

```bash theme={null}
curl -X POST https://api.geekhub.mx/v1/audio/speech \
  -H "Authorization: Bearer ghub_sk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -o speech.mp3 \
  -d '{
    "model": "openai/tts-1",
    "input": "Hello, this is a voice synthesis test.",
    "voice": "alloy",
    "response_format": "mp3",
    "speed": 1.0
  }'
```

Response is **binary audio directly** (not JSON). `Content-Type` indicates the format.

## Output formats

| `response_format` | Content-Type | When to use                                  |
| ----------------- | ------------ | -------------------------------------------- |
| `mp3` (default)   | `audio/mpeg` | Compatible with all browsers and players     |
| `opus`            | `audio/opus` | Low-latency bidirectional streaming (WebRTC) |
| `wav`             | `audio/wav`  | Post-processing (ASR, mixing). Lossless      |
| `pcm`             | `audio/L16`  | No container, custom engine integration      |
| `aac`             | `audio/aac`  | Native iOS / Apple devices                   |
| `flac`            | `audio/flac` | Lossless, compressed. Archival               |

## Real-time streaming

Pass `stream: true` to receive audio in chunks as the model generates it (lower perceptual latency):

```bash theme={null}
curl -N -X POST https://api.geekhub.mx/v1/audio/speech \
  -H "Authorization: Bearer ghub_sk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/tts-1",
    "input": "Text to play while it generates...",
    "voice": "nova",
    "response_format": "opus",
    "stream": true
  }' > stream.opus
```

## Tone instructions (gpt-4o-mini-tts)

The `openai/gpt-4o-mini-tts` model accepts an `instructions` field with style prompting:

```json theme={null}
{
  "model": "openai/gpt-4o-mini-tts",
  "input": "Welcome to Geek Hub.",
  "voice": "verse",
  "response_format": "mp3",
  "instructions": "Speak with a warm, professional tone, paced and clear."
}
```

## Available models

| Model ID                 | Voices                                         | Price           | ZDR |
| ------------------------ | ---------------------------------------------- | --------------- | --- |
| `openai/tts-1`           | alloy, echo, fable, onyx, nova, shimmer        | \$15 / 1M chars | ✓   |
| `openai/tts-1-hd`        | alloy, echo, fable, onyx, nova, shimmer        | \$30 / 1M chars | ✓   |
| `openai/gpt-4o-mini-tts` | alloy, echo, fable, onyx, nova, shimmer, verse | \$12 / 1M chars | ✓   |

## Pricing structure

TTS models are charged per **characters of synthesized text**, not tokens or seconds. This mirrors provider pricing and makes cost predictable: 1000 characters ≈ 150 words regardless of audio duration.

Geek Hub applies the standard +5% markup.

## ZDR pre-flight

If your org requires ZDR for the TTS model's group, or if the request includes `zdr: true`, the gateway verifies before processing. Unverified → HTTP 422 with alternative models.

See [Zero Data Retention](/en/features/zdr).