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

# TTS Voices and Audio Quality Settings in Origami AI

> Configure Kokoro.js in-browser neural TTS voices, accents, and quantization quality for narration, Shorts voiceovers, and Voice Studio.

Origami AI uses [Kokoro.js](https://github.com/Kokoro-js) for all speech synthesis—a fully in-browser neural TTS engine that runs on your device with no server round-trip. Voice generation is private, fast, and works offline after the initial model download. You can choose from dozens of voices spanning multiple accents and languages, each with its own pitch and delivery style.

***

## Choosing a Voice

Origami AI lets you set voices at four different levels, from global defaults to per-slide overrides.

**Settings → General — Default PDF-to-video voice**

The voice you select here is used for all slides when you generate narration audio on a new project. It becomes your studio default.

**Slide Editor — Per-slide voice override**

Inside the Slide Editor, open the audio settings for any individual slide to assign a different voice to that slide only. This is useful for emphasis, character differentiation, or accessibility.

**Shorts Composer — Shorts TTS voice**

The Shorts composer has its own voice setting that controls narration across all scenes in a Short. Set it in the Shorts composer panel or in **⚙️ Settings → General** under the Shorts section.

**Voice Studio — Standalone TTS**

The Voice Studio lets you type any text and synthesize it with a chosen voice, speed, and MP3 bitrate. Set the default voice for Voice Studio sessions in **⚙️ Settings → General** under the Voice Studio section.

***

## Voice Naming System

Every Kokoro.js voice follows a two-letter prefix that tells you the accent and gender at a glance.

### American English

| Prefix | Description     | Example Voices                       |
| ------ | --------------- | ------------------------------------ |
| `af_`  | American Female | `af_heart`, `af_bella`, `af_jessica` |
| `am_`  | American Male   | `am_adam`, `am_michael`, `am_echo`   |

### British English

| Prefix | Description    | Example Voices           |
| ------ | -------------- | ------------------------ |
| `bf_`  | British Female | `bf_emma`, `bf_alice`    |
| `bm_`  | British Male   | `bm_daniel`, `bm_george` |

### European Languages

| Prefix | Description     |
| ------ | --------------- |
| `ef_`  | European Female |
| `em_`  | European Male   |
| `ff_`  | French Female   |
| `if_`  | Italian Female  |
| `im_`  | Italian Male    |

### Asian Languages

| Prefix | Description     |
| ------ | --------------- |
| `jf_`  | Japanese Female |
| `jm_`  | Japanese Male   |
| `zf_`  | Chinese Female  |
| `zm_`  | Chinese Male    |

<Note>
  Voice availability may vary depending on your installed Kokoro.js model version. If a voice you've selected stops producing audio, switch to another voice in the same prefix group—they share the same accent and gender characteristics.
</Note>

***

## Quality Settings

Open **⚙️ Settings → TTS Model** to control audio fidelity and normalization behavior.

### Quantization

Quantization determines the numerical precision used during TTS inference, trading off audio quality against speed and model size.

| Setting | Quality                                      | Use When                             |
| ------- | -------------------------------------------- | ------------------------------------ |
| **q8**  | High quality, slightly larger model download | Final exports, published videos      |
| **q4**  | Faster generation, acceptable quality        | Quick previews, iterating on scripts |

<Tip>
  Use **q8** when you're ready to export your final video and **q4** while you're still writing and previewing scripts. The difference in generation time is noticeable on longer projects.
</Tip>

### Disable Audio Normalization

By default, Origami AI normalizes each audio clip so narration volumes stay consistent across slides. If the automatic normalization makes voices sound unnatural—particularly with whisper-style or low-dynamic-range voices—toggle **Disable Audio Normalization** to preserve the raw output from Kokoro.js.