tattn/voicevox-client

CLI

Install with Homebrew

brew tap tattn/voicevox-client
brew install voicevox-client

Setup

voicevox-client setup

Resources are saved to ~/.voicevox-client/resources/.

Set VOICEVOX_CLIENT_HOME to change the base directory (default: ~/.voicevox-client).

Usage

# Synthesize speech (uses default resources)
voicevox-client --text "こんにちは"

# Specify style and output
voicevox-client --text "こんにちは" --style-id 1 --output greeting.wav

# Adjust speech parameters
voicevox-client --text "こんにちは" --speed 1.2 --pitch 0.8

# AquesTalk-like kana notation (control accent directly)
voicevox-client --text "コンニチワ'" --kana

# Use a specific voice model by name (loads <resources>/vvms/<name>.vvm)
voicevox-client --text "こんにちは" --model-name 1

# List speakers in the current model
voicevox-client speakers

# List all speakers from all downloaded models
voicevox-client speakers --all

<details> <summary>AquesTalk-like kana notation reference</summary>

The --kana flag accepts katakana with special symbols to control accent and intonation:

| Symbol | Description | Example | |---|---|---| | ' | Accent position (required once per phrase) | コンニチワ' | | / | Accent phrase separator | コンニチワ'/ゲンキデ'スカ | | 、 | Accent phrase separator with pause | コンニチワ'、ゲンキデ'スカ | | | Unvoiced mora (place before kana) | スキデ'ス | | ？ | Interrogative intonation (place at end) | ゲンキデ'スカ？ |

</details>

<details> <summary>Audio query (JSON) workflow</summary>

You can export an audio query as JSON, edit it, and then synthesize from it:

# Generate audio query JSON
voicevox-client query --text "こんにちは" --style-id 0 > query.json

# Save to a file directly
voicevox-client query --text "こんにちは" --style-id 0 --output query.json

# Synthesize from a JSON file
voicevox-client synthesize-from-query --input query.json --style-id 0 --output output.wav

# Pipe directly
voicevox-client query --text "こんにちは" | voicevox-client synthesize-from-query --input - --output output.wav

</details>

Swift Library

The API documentation is available here.

import VOICEVOX

let config = VOICEVOXConfiguration(
    openJTalkDictionaryURL: Bundle.main.url(forResource: "open_jtalk_dic", withExtension: nil)!
)

let synthesizer = try await Synthesizer(configuration: config)
try await synthesizer.loadVoiceModel(from: modelURL)

let audioData = try await synthesizer.synthesize(text: "こんにちは", styleId: 0)

// AquesTalk-like kana notation for direct accent control
let audioData = try await synthesizer.synthesize(kana: "コンニチワ'", styleId: 0)

Example App Setup

./scripts/setup-voicevox-resources.sh

This downloads VOICEVOX Core, ONNX Runtime, OpenJTalk dictionary, and voice models to Example/VOICEVOXExample/lib/. If you hit GitHub release rate limits, set GITHUB_TOKEN or GH_TOKEN before running it.

Package Metadata

Repository: tattn/voicevox-client

Default branch: main

README: README.md