---
title: Gemini 3.8 Flash TTS | Runware Docs
url: https://runware.ai/docs/models/google-gemini-3-8-flash-tts
description: Directable text-to-speech with style control, two-speaker dialogue, and 130+ languages
---
# Gemini 3.8 Flash TTS

![Gemini 3.8 Flash TTS](https://assets.runware.ai/covers/google-gemini-3-8-flash-tts.jpg)

Gemini 3.8 Flash TTS is Google's text-to-speech model for expressive, natural-sounding speech from text. It follows natural-language delivery directions for tone, pace, and emotion, renders inline tags such as laughs, sighs, and pauses, and generates two-speaker dialogue in a single request. It detects the language of the text automatically across more than 130 languages, making it suitable for narration, voice agents, podcasts, audiobooks, and localized voiceover.

- **Status**: live
- **ID**: `google:gemini@3.8-flash-tts`
- Try in Playground
- **Creator**: Google
- **Release Date**: September 22, 2026
- **Zero Data Retention**: Yes
- **Capabilities**: Text to Audio, Checkpoint

## Pricing

~$0.0333 typical run

**Rates** (per 1M tokens)

- Input: `$1.00`
- Output: `$18.00`

**4 example generations**

- [multi-speaker-dialogue](https://runware.ai/docs/models/google-gemini-3-8-flash-tts/examples#example-backcountry-forecast-segment-stinger): `$0.00668` (~8s)
- [text-to-audio](https://runware.ai/docs/models/google-gemini-3-8-flash-tts/examples#example-coral-cryobank-fundraising-voiceover): `$0.0258` (~15s)
- [text-to-audio](https://runware.ai/docs/models/google-gemini-3-8-flash-tts/examples#example-forensic-timber-podcast-cold-open): `$0.0407` (~23s)
- [multi-speaker-dialogue](https://runware.ai/docs/models/google-gemini-3-8-flash-tts/examples#example-flood-damaged-archive-training-dialogue): `$0.0527` (~14s)

billed per token, so a request costs what the model produced · times vary, measured end to end · [How pricing works](https://runware.ai/docs/models-api/pricing)

## Compatibility & Validation

Inside `speech`, `voice` cannot be used with `voices`.

## Request Parameters

**API Options**

Platform-level options for task execution and delivery.

### [taskType](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-tasktype)

- **Type**: `string`
- **Required**: true
- **Value**: `audioInference`

Identifier for the type of task being performed.

### [taskUUID](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-taskuuid)

- **Type**: `string`
- **Required**: true
- **Format**: `UUID v4`

UUID v4 identifier for tracking tasks and matching async responses. Must be unique per task.

### [outputType](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-outputtype)

- **Type**: `string`
- **Default**: `URL`

Audio output type.

**Allowed values**: `URL` `base64Data` `dataURI`

### [outputFormat](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-outputformat)

- **Type**: `string`
- **Default**: `MP3`

Specifies the file format of the generated output. The available values depend on the task type and the specific model's capabilities.

- \`MP3\`: Compressed audio, smaller file size.
- \`WAV\`: Uncompressed, high-quality audio.
- \`FLAC\`: Lossless compression.
- \`OGG\`: Open-source compressed audio format (Vorbis codec).

**Allowed values**: `MP3` `WAV` `FLAC` `OGG`

### [audioSettings](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-audiosettings)

- **Path**: `audioSettings.bitrate`
- **Type**: `object (3 properties)`

Audio encoding settings for controlling the bitrate, number of channels, and sample rate of the generated audio. Only applicable for lossy output formats (`MP3` and `OGG`). When using lossless formats (`WAV` or `FLAC`), this parameter must not be provided.

The available sample rates and valid bitrate ranges depend on the output format. For `OGG`, bitrate limits also vary by the number of channels.

**MP3 bitrate limits**:

Bitrate limits for MP3 are the same regardless of mono or stereo.

| Sample Rate | Min Bitrate | Max Bitrate |
| --- | --- | --- |
| 8,000 Hz | 8 kbps | 64 kbps |
| 11,025 Hz | 8 kbps | 64 kbps |
| 12,000 Hz | 8 kbps | 64 kbps |
| 16,000 Hz | 8 kbps | 160 kbps |
| 22,050 Hz | 8 kbps | 160 kbps |
| 24,000 Hz | 8 kbps | 160 kbps |
| 32,000 Hz | 32 kbps | 320 kbps |
| 44,100 Hz | 32 kbps | 320 kbps |
| 48,000 Hz | 32 kbps | 320 kbps |

**OGG bitrate limits — Mono (1 channel)**:

| Sample Rate | Min Bitrate | Max Bitrate |
| --- | --- | --- |
| 8,000 Hz | 8 kbps | 40 kbps |
| 12,000 Hz | 16 kbps | 48 kbps |
| 16,000 Hz | 16 kbps | 96 kbps |
| 24,000 Hz | 16 kbps | 80 kbps |
| 48,000 Hz | 32 kbps | 224 kbps |

**OGG bitrate limits — Stereo (2 channels)**:

| Sample Rate | Min Bitrate | Max Bitrate |
| --- | --- | --- |
| 8,000 Hz | 16 kbps | 80 kbps |
| 12,000 Hz | 16 kbps | 96 kbps |
| 16,000 Hz | 24 kbps | 192 kbps |
| 24,000 Hz | 32 kbps | 160 kbps |
| 48,000 Hz | 48 kbps | 256 kbps |

> [!NOTE]
> **Lossless formats**: When `outputFormat` is set to `WAV` or `FLAC`, the `audioSettings` parameter is not available since these formats produce uncompressed or lossless audio with no configurable encoding settings.

#### [bitrate](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-audiosettings-bitrate)

- **Path**: `audioSettings.bitrate`
- **Type**: `integer`
- **Min**: `8`

Audio bitrate in kbps.

#### [channels](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-audiosettings-channels)

- **Path**: `audioSettings.channels`
- **Type**: `integer`
- **Default**: `2`

Number of audio channels. 1 for mono, 2 for stereo.

**Allowed values**: `1` `2`

#### [sampleRate](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-audiosettings-samplerate)

- **Path**: `audioSettings.sampleRate`
- **Type**: `integer`

Audio sample rate in Hz.

### [webhookURL](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-webhookurl)

- **Type**: `string`
- **Format**: `uri`

Webhook URL that receives JSON responses via HTTP POST when generation tasks complete. For batch requests with multiple results, each completed item triggers a separate webhook call as it becomes available.

**Learn more** (1 resource):

- [Webhooks](https://runware.ai/docs/models-api/webhooks) (platform)

### [deliveryMethod](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-deliverymethod)

- **Type**: `string`
- **Default**: `sync`

Determines how the API delivers task results.

**Allowed values**:

- `sync` Returns results directly in the API response. Over HTTP, a task still running after 120 seconds continues as an async task and comes back with status 'processing'. Poll for its results using getResponse.
- `async` Returns an immediate acknowledgment with the task UUID. Poll for results using getResponse.

**Learn more** (1 resource):

- [Task Polling](https://runware.ai/docs/models-api/task-polling) (platform)

### [uploadEndpoint](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-uploadendpoint)

- **Type**: `string`
- **Format**: `uri`

Specifies a URL where the generated content will be automatically uploaded using the HTTP PUT method. The raw binary data of the media file is sent directly as the request body. For secure uploads to cloud storage, use presigned URLs that include temporary authentication credentials.

**Common use cases:**

- **Cloud storage**: Upload directly to S3 buckets, Google Cloud Storage, or Azure Blob Storage using presigned URLs.
- **CDN integration**: Upload to content delivery networks for immediate distribution.

```text
// S3 presigned URL for secure upload
https://your-bucket.s3.amazonaws.com/generated/content.mp4?X-Amz-Signature=abc123&X-Amz-Expires=3600

// Google Cloud Storage presigned URL
https://storage.googleapis.com/your-bucket/content.jpg?X-Goog-Signature=xyz789

// Custom storage endpoint
https://storage.example.com/uploads/generated-image.jpg
```

The content data will be sent as the request body to the specified URL when generation is complete.

### [ttl](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-ttl)

- **Type**: `integer`
- **Min**: `60`

Time-to-live (TTL) in seconds for generated content. Only applies when `outputType` is `URL`.

### [includeCost](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-includecost)

- **Type**: `boolean`

Include task cost in the response.

### [numberResults](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-numberresults)

- **Type**: `integer`
- **Min**: `1`
- **Max**: `4`
- **Default**: `1`

Number of results to generate. Each result uses a different seed, producing variations of the same parameters.

**Core Parameters**

Primary parameters that define the task output.

### [model](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-model)

- **Type**: `string`
- **Required**: true
- **Value**: `google:gemini@3.8-flash-tts`

Send `auto` instead of an AIR identifier to have the model router choose the model for this task. The rest of the request is unchanged, every value you send is kept, and only the parameters you leave out are filled in. See [Model Router](https://runware.ai/docs/models-api/model-router).

**Learn more** (3 resources):

- [Text To Image: Model Selection](https://runware.ai/docs/learn/text-to-image#model-selection) (learn)
- [Image Inpainting: Model Specialized Inpainting Models](https://runware.ai/docs/learn/image-inpainting#model-specialized-inpainting-models) (learn)
- [Image Outpainting: Other Critical Parameters](https://runware.ai/docs/learn/image-outpainting#other-critical-parameters) (learn)

### [speech](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech)

- **Path**: `speech.text`
- **Type**: `object (5 properties)`
- **Required**: true

Settings for speech generation.

#### [text](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech-text)

- **Path**: `speech.text`
- **Type**: `string`
- **Required**: true
- **Min**: `1`
- **Max**: `4000`

Text to convert to speech. Inline tags such as `<laugh>`, `<sigh>` and `<short pause>` add sounds and pauses. In dialogue mode, each turn starts with a speaker tag such as `[Sam]`.

#### [voice](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech-voice)

- **Path**: `speech.voice`
- **Type**: `string`
- **Default**: `Zephyr`

Voice identifier to use for speech generation.

**Allowed values**:

- `Achernar` Female, Soft
- `Achird` Male, Friendly
- `Algenib` Male, Gravelly
- `Algieba` Male, Smooth
- `Alnilam` Male, Firm
- `Aoede` Female, Breezy
- `Autonoe` Female, Bright
- `Callirrhoe` Female, Easy-going
- `Charon` Male, Informative
- `Despina` Female, Smooth
- `Enceladus` Male, Breathy
- `Erinome` Female, Clear
- `Fenrir` Male, Excitable
- `Gacrux` Female, Mature
- `Iapetus` Male, Clear
- `Kore` Female, Firm
- `Laomedeia` Female, Upbeat
- `Leda` Female, Youthful
- `Orus` Male, Firm
- `Puck` Male, Upbeat
- `Pulcherrima` Female, Forward
- `Rasalgethi` Male, Informative
- `Sadachbia` Male, Lively
- `Sadaltager` Male, Knowledgeable
- `Schedar` Male, Even
- `Sulafat` Female, Warm
- `Umbriel` Male, Easy-going
- `Vindemiatrix` Female, Gentle
- `Zephyr` Female, Bright
- `Zubenelgenubi` Male, Casual

#### [voices](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech-voices)

- **Path**: `speech.voices`
- **Type**: `array of objects (2 properties)`
- **Min items**: `items: 2`

Two-speaker dialogue configuration.

##### [speaker](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech-voices-speaker)

- **Path**: `speech.voices.speaker`
- **Type**: `string`
- **Required**: true
- **Min**: `1`

Unique speaker alias. Must appear in the text as `[Alias]`.

##### [voice](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-speech-voices-voice)

- **Path**: `speech.voices.voice`
- **Type**: `string`
- **Required**: true
- **Default**: `Zephyr`

Voice identifier to use for speech generation.

**Allowed values**:

- `Achernar` Female, Soft
- `Achird` Male, Friendly
- `Algenib` Male, Gravelly
- `Algieba` Male, Smooth
- `Alnilam` Male, Firm
- `Aoede` Female, Breezy
- `Autonoe` Female, Bright
- `Callirrhoe` Female, Easy-going
- `Charon` Male, Informative
- `Despina` Female, Smooth
- `Enceladus` Male, Breathy
- `Erinome` Female, Clear
- `Fenrir` Male, Excitable
- `Gacrux` Female, Mature
- `Iapetus` Male, Clear
- `Kore` Female, Firm
- `Laomedeia` Female, Upbeat
- `Leda` Female, Youthful
- `Orus` Male, Firm
- `Puck` Male, Upbeat
- `Pulcherrima` Female, Forward
- `Rasalgethi` Male, Informative
- `Sadachbia` Male, Lively
- `Sadaltager` Male, Knowledgeable
- `Schedar` Male, Even
- `Sulafat` Female, Warm
- `Umbriel` Male, Easy-going
- `Vindemiatrix` Female, Gentle
- `Zephyr` Female, Bright
- `Zubenelgenubi` Male, Casual

**Settings**

Technical parameters to fine-tune the inference process. These must be nested inside the \`settings\` object.

### [style](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#request-settings-style)

- **Path**: `settings.style`
- **Type**: `string`
- **Min**: `1`

Delivery direction for the speech, such as tone, pace or emotion. Applies to every turn.

## Response Parameters

### [taskType](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-tasktype)

- **Type**: `string`
- **Required**: true
- **Value**: `audioInference`

Identifier for the type of task this response belongs to.

### [taskUUID](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-taskuuid)

- **Type**: `string`
- **Required**: true
- **Format**: `UUID v4`

UUID v4 identifier echoed from the original request, used to match async responses to their tasks.

### [audioUUID](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-audiouuid)

- **Type**: `string`
- **Required**: true
- **Format**: `UUID v4`

UUID of the output audio.

### [audioURL](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-audiourl)

- **Type**: `string`
- **Format**: `uri`

URL of the output audio.

### [audioBase64Data](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-audiobase64data)

- **Type**: `string`

Base64-encoded audio data.

### [audioDataURI](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-audiodatauri)

- **Type**: `string`
- **Format**: `uri`

Data URI of the output audio.

### [cost](https://runware.ai/docs/models/google-gemini-3-8-flash-tts#response-cost)

- **Type**: `float`

Task cost in USD. Present when `includeCost` is set to `true` in the request.

## Examples

### text-to-audio (text-to-audio)

[Listen to audio](https://assets.runware.ai/examples/google-gemini-3-8-flash-tts/ccdc6cbd-57ab-4389-89c5-904c9c82aeb9.mp3)

> **Prompt**: At first, it looks like a snow globe: tiny coral cells turning slowly in a vial. But this is not a souvenir. It is a future reef. At minus one hundred and ninety-six degrees Celsius, coral reproductive cells can be preserved for the decades ahead—protecting genetic diversity that warming seas could erase in a single season. <short pause> When conditions recover, scientists can thaw viable samples and give restoration teams more choices: corals that may tolerate heat, resist disease, and help damaged reefs begin again. One vial can safeguard possibilities we have not even discovered yet. <short pause> Support coral cryobanking, and you are not simply freezing life in time. You are keeping the door open—for bright shallows, crowded nurseries, and reefs that still have a chance to surprise us.

```json
{
  "taskType": "audioInference",
  "taskUUID": "f6694623-b5e6-459d-a3d9-b27eda4f88fe",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "At first, it looks like a snow globe: tiny coral cells turning slowly in a vial. But this is not a souvenir. It is a future reef.\n\nAt minus one hundred and ninety-six degrees Celsius, coral reproductive cells can be preserved for the decades ahead—protecting genetic diversity that warming seas could erase in a single season. <short pause>\n\nWhen conditions recover, scientists can thaw viable samples and give restoration teams more choices: corals that may tolerate heat, resist disease, and help damaged reefs begin again.\n\nOne vial can safeguard possibilities we have not even discovered yet. <short pause>\n\nSupport coral cryobanking, and you are not simply freezing life in time. You are keeping the door open—for bright shallows, crowded nurseries, and reefs that still have a chance to surprise us.",
    "voice": "Sulafat"
  },
  "settings": {
    "style": "Warm, intimate documentary narration with quiet wonder at the opening, measured clarity through the scientific explanation, and a gradual rise into sincere, hopeful conviction; use natural pauses and avoid a conventional sales-announcer tone."
  }
}
```

---

### text-to-audio (text-to-audio)

[Listen to audio](https://assets.runware.ai/examples/google-gemini-3-8-flash-tts/38d49973-7065-4507-8fbc-4543007d20f7.mp3)

> **Prompt**: A shipping container arrives at the Port of Rotterdam carrying thirty tons of tropical hardwood. The paperwork says plantation-grown teak. Legal. Traceable. Unremarkable. But inside a government laboratory, Dr. Lena Vos lifts one splinter from the shipment and places it beneath a mass spectrometer. <short pause> The machine finds something the paperwork forgot to mention: rain. Not last week’s rain, but decades of it—stored in the wood as a chemical signature. The pattern points thousands of kilometers away from the declared plantation, toward a protected forest where logging is forbidden. That splinter is now evidence. <short pause> We tend to imagine timber crime happening deep in the jungle: chainsaws, trucks, men working by moonlight. But the decisive moment may come months later, in a silent lab, when a scientist asks a piece of wood where it grew. And, remarkably, the wood answers. I’m Mara Ellison, and this is “Material Witness,” the podcast about ordinary substances that expose extraordinary crimes. Today: the hidden geography inside a tree—and the forensic race to map the world’s forests before they disappear. <sigh> Because once the last tree is cut, even perfect evidence arrives too late.

```json
{
  "taskType": "audioInference",
  "taskUUID": "b0bd0ac5-f325-4a14-876d-0b62c07bd09a",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "A shipping container arrives at the Port of Rotterdam carrying thirty tons of tropical hardwood. The paperwork says plantation-grown teak. Legal. Traceable. Unremarkable.\n\nBut inside a government laboratory, Dr. Lena Vos lifts one splinter from the shipment and places it beneath a mass spectrometer. <short pause> The machine finds something the paperwork forgot to mention: rain. Not last week’s rain, but decades of it—stored in the wood as a chemical signature. The pattern points thousands of kilometers away from the declared plantation, toward a protected forest where logging is forbidden.\n\nThat splinter is now evidence.\n\n<short pause> We tend to imagine timber crime happening deep in the jungle: chainsaws, trucks, men working by moonlight. But the decisive moment may come months later, in a silent lab, when a scientist asks a piece of wood where it grew.\n\nAnd, remarkably, the wood answers.\n\nI’m Mara Ellison, and this is “Material Witness,” the podcast about ordinary substances that expose extraordinary crimes. Today: the hidden geography inside a tree—and the forensic race to map the world’s forests before they disappear. <sigh> Because once the last tree is cut, even perfect evidence arrives too late.",
    "voice": "Gacrux"
  },
  "settings": {
    "style": "Deliver as an intimate, cinematic investigative-podcast host: measured and quietly urgent, with crisp factual authority, restrained emotion, and deliberate pauses that let each revelation land."
  }
}
```

---

### multi-speaker-dialogue (multi-speaker-dialogue)

[Listen to audio](https://assets.runware.ai/examples/google-gemini-3-8-flash-tts/67abfb98-3af3-436f-8e07-9f9fd8002772.mp3)

> **Prompt**: [Nia] Beacon checked? [Owen] Checked. Probe and shovel? [Nia] Packed. Now, the forecast. [Owen] New snow, rising wind, changing risk. [Nia] So before we chase fresh tracks— [Owen] —we read the mountain. <short pause> [Nia] This is the Backcountry Conditions Report.

```json
{
  "taskType": "audioInference",
  "taskUUID": "491ec7f3-2737-463f-bc6e-a1a9ad00fdaf",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "[Nia] Beacon checked?\n[Owen] Checked. Probe and shovel?\n[Nia] Packed. Now, the forecast.\n[Owen] New snow, rising wind, changing risk.\n[Nia] So before we chase fresh tracks—\n[Owen] —we read the mountain. <short pause>\n[Nia] This is the Backcountry Conditions Report.",
    "voices": [
      {
        "speaker": "Nia",
        "voice": "Kore"
      },
      {
        "speaker": "Owen",
        "voice": "Algieba"
      }
    ]
  },
  "settings": {
    "style": "Deliver as a polished podcast segment stinger: crisp, confident, safety-minded, and naturally conversational, with brisk pacing and subtle rhythmic interplay rather than an announcer voice."
  }
}
```

---

### multi-speaker-dialogue (multi-speaker-dialogue)

[Listen to audio](https://assets.runware.ai/examples/google-gemini-3-8-flash-tts/1d278ee4-98f0-4e38-a257-75285aed5e7c.mp3)

> **Prompt**: [Mara] The sprinkler line has stopped, but these records are still absorbing water. Eli, what do you do first? [Eli] Keep everyone out until facilities confirms the area is electrically safe. Then I put on nitrile gloves, photograph the room, and record where each box was found. [Mara] Good. Documentation preserves context. Now you have three groups: damp folders, soaked bound volumes, and glossy photographs stuck together. Which group gets immediate attention? [Eli] The photographs. If their emulsions dry while touching, they may bond permanently. I keep them wet in clean, cool water until a conservator can separate them. [Mara] Exactly. Never peel them apart. What about the damp folders? [Eli] If there are only a few, I move them to a clean, ventilated space. I stand folders upright, fan the pages gently, and place absorbent paper beneath them. I change that paper as it becomes wet. [Mara] And no direct heat. It can distort paper and accelerate damage. <short pause> The soaked volumes are too numerous to air-dry within forty-eight hours. Your next step? [Eli] Pack them for freezing. I wrap each volume in freezer paper, spine down, without forcing it open. Then I label the package with its collection and box number. [Mara] Why the forty-eight-hour deadline? [Eli] Mold can begin growing quickly in warm, humid conditions. Freezing buys time and stabilizes the material until vacuum freeze-drying is available. [Mara] One last check: you notice a gray, fuzzy patch on a carton. [Eli] I stop handling it, isolate the carton, and notify the response lead. I do not brush or vacuum suspected mold without the right containment and protective equipment. [Mara] Correct. In a records emergency, the priorities are safety, documentation, stabilization, and traceable decisions—not speed at any cost.

```json
{
  "taskType": "audioInference",
  "taskUUID": "efc1772e-201b-4905-9846-996529d8212a",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "[Mara] The sprinkler line has stopped, but these records are still absorbing water. Eli, what do you do first?\n[Eli] Keep everyone out until facilities confirms the area is electrically safe. Then I put on nitrile gloves, photograph the room, and record where each box was found.\n[Mara] Good. Documentation preserves context. Now you have three groups: damp folders, soaked bound volumes, and glossy photographs stuck together. Which group gets immediate attention?\n[Eli] The photographs. If their emulsions dry while touching, they may bond permanently. I keep them wet in clean, cool water until a conservator can separate them.\n[Mara] Exactly. Never peel them apart. What about the damp folders?\n[Eli] If there are only a few, I move them to a clean, ventilated space. I stand folders upright, fan the pages gently, and place absorbent paper beneath them. I change that paper as it becomes wet.\n[Mara] And no direct heat. It can distort paper and accelerate damage. <short pause> The soaked volumes are too numerous to air-dry within forty-eight hours. Your next step?\n[Eli] Pack them for freezing. I wrap each volume in freezer paper, spine down, without forcing it open. Then I label the package with its collection and box number.\n[Mara] Why the forty-eight-hour deadline?\n[Eli] Mold can begin growing quickly in warm, humid conditions. Freezing buys time and stabilizes the material until vacuum freeze-drying is available.\n[Mara] One last check: you notice a gray, fuzzy patch on a carton.\n[Eli] I stop handling it, isolate the carton, and notify the response lead. I do not brush or vacuum suspected mold without the right containment and protective equipment.\n[Mara] Correct. In a records emergency, the priorities are safety, documentation, stabilization, and traceable decisions—not speed at any cost.",
    "voices": [
      {
        "speaker": "Mara",
        "voice": "Gacrux"
      },
      {
        "speaker": "Eli",
        "voice": "Charon"
      }
    ]
  },
  "settings": {
    "style": "Deliver as a calm, realistic workplace training conversation: measured pace, clear instructional emphasis, and understated urgency without sounding theatrical."
  }
}
```