API Reference - VOICEVOX Core for PHP

`Onnxruntime`

ONNX Runtime loader. This is a process-level singleton — only one instance exists per process.

Methods

Method	Description
`static loadOnce(string $filename = ''): self`	Loads and initializes the ONNX Runtime. On subsequent calls the argument is ignored and the existing instance is returned.
`static get(): ?self`	Returns the existing instance, or `null` if not yet initialized.
`supportedDevices(): string`	Returns available device information as a JSON string.
`static libRecommendedVersionedFilename(): string`	Returns the recommended versioned filename of the ONNX Runtime library (e.g., `libvoicevox_onnxruntime.1.23.2.dylib`).
`static libRecommendedUnversionedFilename(): string`	Returns the recommended unversioned filename of the ONNX Runtime library.
`static libMinRequiredMinorVersion(): int`	Returns the minimum supported ONNX Runtime 1.x minor version.
`static libMaxSupportedMinorVersion(): int`	Returns the maximum supported ONNX Runtime 1.x minor version.

libVersionedFilename() and libUnversionedFilename() remain as deprecated aliases.

Constants

Constant	Description
`LIB_MIN_REQUIRED_MINOR_VERSION`	Minimum supported ONNX Runtime 1.x minor version (`17`)
`LIB_MAX_SUPPORTED_MINOR_VERSION`	Maximum supported ONNX Runtime 1.x minor version (`29`)
`LIB_RECOMMENDED_NAME`	Recommended library base name (`voicevox_onnxruntime`)
`LIB_RECOMMENDED_VERSION`	Recommended ONNX Runtime version (`1.23.2`)

LIB_NAME and LIB_VERSION remain as deprecated aliases.

`OpenJtalk`

Text analyzer using OpenJTalk.

Methods

Method	Description
`__construct(string $openJtalkDictDir)`	Initializes with the path to the OpenJTalk dictionary directory.
`analyze(string $text): string`	Analyzes Japanese text and returns an accent phrase array as a JSON string.
`useUserDict(UserDict $userDict): void`	Applies a user dictionary. Call again if the dictionary is updated.

`VoiceModelFile`

Voice model file (.vvm file).

Methods

Method	Description
`static open(string $path): self`	Opens a `.vvm` file.
`id(): string`	Returns the voice model ID as a hex string (16 bytes).
`createMetasJson(): string`	Returns speaker metadata as a JSON string.
`close(): void`	Closes the file and releases resources.

`Synthesizer`

Main class for text-to-speech synthesis.

Methods

Method	Description
`__construct(Onnxruntime $onnxruntime, OpenJtalk $openJtalk, AccelerationMode $accelerationMode = Auto, int $cpuNumThreads = 0)`	Initializes the synthesizer.
`onnxruntime(): Onnxruntime`	Returns the `Onnxruntime` instance held by this synthesizer.
`isGpuMode(): bool`	Returns whether GPU mode is enabled.
`metas(): string`	Returns metadata for all loaded speakers as a JSON string.
`loadVoiceModel(VoiceModelFile $model, OnExistingVoiceModelId $onExisting = Error): void`	Loads a voice model. Use `$onExisting` to choose what happens when a model with the same ID is already loaded.
`unloadVoiceModel(string $voiceModelId): void`	Unloads a voice model by its hex ID.
`isLoadedVoiceModel(string $voiceModelId): bool`	Returns whether a voice model is loaded.
`createAudioQuery(string $text, int $styleId): string`	Generates an AudioQuery JSON from Japanese text.
`createAudioQueryFromKana(string $kana, int $styleId): string`	Generates an AudioQuery JSON from AquesTalk-style kana notation.
`createAccentPhrases(string $text, int $styleId): string`	Generates an accent phrase array JSON from Japanese text.
`createAccentPhrasesFromKana(string $kana, int $styleId): string`	Generates an accent phrase array JSON from kana notation.
`replaceMoraData(string $accentPhrasesJson, int $styleId): string`	Returns updated accent phrases with mora pitch and phoneme length from another style.
`replacePhonemeLength(string $accentPhrasesJson, int $styleId): string`	Returns updated accent phrases with phoneme lengths from another style.
`replaceMoraPitch(string $accentPhrasesJson, int $styleId): string`	Returns updated accent phrases with mora pitch from another style.
`synthesis(string $audioQueryJson, int $styleId, bool $enableInterrogativeUpspeak = true): string`	Synthesizes speech from an AudioQuery JSON. Returns WAV binary.
`createAudioFeature(string $audioQueryJson, int $styleId, bool $enableInterrogativeUpspeak = true): AudioFeature`	Creates the `AudioFeature` intermediate representation used for frame-based rendering.
`tts(string $text, int $styleId, bool $enableInterrogativeUpspeak = true): string`	Synthesizes speech from Japanese text in one step. Returns WAV binary.
`ttsFromKana(string $kana, int $styleId, bool $enableInterrogativeUpspeak = true): string`	Synthesizes speech from kana notation. Returns WAV binary.
`createSingFrameAudioQuery(string $scoreJson, int $styleId): string`	Generates a singing synthesis query JSON from a score JSON.
`frameSynthesis(string $frameAudioQueryJson, int $styleId): string`	Synthesizes singing audio from a frame audio query. Returns WAV binary.
`createSingFrameF0(string $scoreJson, string $frameAudioQueryJson, int $styleId): string`	Generates per-frame F0 (fundamental frequency) values as a JSON float array.
`createSingFrameVolume(string $scoreJson, string $frameAudioQueryJson, int $styleId): string`	Generates per-frame volume values as a JSON float array.

`AudioFeature` and `AudioFeatureSlice`

AudioFeature owns the native intermediate representation, and each AudioFeatureSlice keeps its owning AudioFeature alive. Rendering is synchronous and returns raw 16-bit PCM bytes.

Class / method	Description
`AudioFeature::length(): int`	Returns the total number of frames.
`AudioFeature::isEmpty(): bool`	Returns whether the feature has no frames.
`AudioFeature::whole(): AudioFeatureSlice`	Returns a slice covering the whole feature.
`AudioFeature::slice(int $startInclusive, int $end, bool $endIsInclusive = false, bool $clamp = false, bool $allowStartExceedsEnd = false): AudioFeatureSlice`	Creates a frame-range slice. The end is exclusive by default.
`AudioFeature::render(): string`	Renders the whole feature as PCM.
`AudioFeatureSlice::length(): int`	Returns the number of frames in the slice.
`AudioFeatureSlice::isEmpty(): bool`	Returns whether the slice has no frames.
`AudioFeatureSlice::slice(int $startInclusive, int $end, bool $endIsInclusive = false, bool $clamp = false, bool $allowStartExceedsEnd = false): AudioFeatureSlice`	Creates a subslice using the same range options.
`AudioFeatureSlice::render(): string`	Renders the slice as PCM.

`VoicevoxCore`

Global utility functions for VOICEVOX Core.

Methods

Method	Description
`getVersion(): string`	Returns the VOICEVOX Core version as a SemVer string.
`getFrameRate(): float`	Returns the number of audio frames per second (`93.75` in CORE 0.17.1).
`audioQueryFrameLength(string $audioQueryJson, bool $enableInterrogativeUpspeak = true): int`	Returns the estimated audio length in frames.
`audioQueryCreateFromAccentPhrases(string $accentPhrasesJson): string`	Generates an AudioQuery JSON from an accent phrase array JSON.
`audioQueryValidate(string $audioQueryJson): void`	Validates an `AudioQuery` JSON. Throws `VoicevoxException` if invalid.
`accentPhraseValidate(string $accentPhraseJson): void`	Validates an `AccentPhrase` JSON. Throws `VoicevoxException` if invalid.
`moraValidate(string $moraJson): void`	Validates a `Mora` JSON. Throws `VoicevoxException` if invalid.
`scoreValidate(string $scoreJson): void`	Validates a `Score` JSON. Throws `VoicevoxException` if invalid.
`noteValidate(string $noteJson): void`	Validates a `Note` JSON. Throws `VoicevoxException` if invalid.
`frameAudioQueryValidate(string $frameAudioQueryJson): void`	Validates a `FrameAudioQuery` JSON. Throws `VoicevoxException` if invalid.
`framePhonemeValidate(string $framePhonemeJson): void`	Validates a `FramePhoneme` JSON. Throws `VoicevoxException` if invalid.
`ensureCompatible(string $scoreJson, string $frameAudioQueryJson): void`	Checks that a score and frame audio query are compatible. Throws `VoicevoxException` if not.

`UserDict`

User dictionary for registering custom word pronunciations.

Methods

Method	Description
`__construct()`	Creates an empty user dictionary.
`load(string $path): void`	Loads a user dictionary from a file.
`save(string $path): void`	Saves the user dictionary to a file.
`addWord(string $surface, string $pronunciation, int $accentType, UserDictWordType $wordType = CommonNoun, int $priority = 5): string`	Adds a word and returns its UUID as a hex string.
`updateWord(string $wordUuid, string $surface, string $pronunciation, int $accentType, UserDictWordType $wordType = CommonNoun, int $priority = 5): void`	Updates an existing word identified by UUID.
`removeWord(string $wordUuid): void`	Removes the word identified by UUID.
`importDict(UserDict $other): void`	Imports words from another `UserDict`.
`toJson(): string`	Returns all words as a JSON string.

$priority must be between 0 and 10. Values outside that range throw an InvalidArgumentException.

Enum: `AccelerationMode`

Hardware acceleration mode for the synthesizer.

Case	Value	Description
`Auto`	`0`	Automatically selects the best available mode.
`Cpu`	`1`	Forces CPU mode.
`Gpu`	`2`	Forces GPU mode.

Enum: `OnExistingVoiceModelId`

Behavior of Synthesizer::loadVoiceModel() when a voice model with the same ID is already loaded.

Case	Value	Description
`Error`	`0`	Throws an error. This is the default.
`Reload`	`1`	Reloads the existing voice model.
`Skip`	`2`	Does nothing.

Enum: `UserDictWordType`

Part-of-speech for user dictionary entries.

Case	Value	Description
`ProperNoun`	`0`	Proper noun
`CommonNoun`	`1`	Common noun
`Verb`	`2`	Verb
`Adjective`	`3`	Adjective
`Suffix`	`4`	Suffix

`VoicevoxException`

Thrown when the VOICEVOX Core C API call returns an error code. The exception message contains the error description from the library.

GitHub Copilot

SDK

AI SDK

VOICEVOX

Starter Kits

Feeds and RSS

Middleware

API Reference - VOICEVOX Core for PHP

`Onnxruntime`

Methods

Constants

`OpenJtalk`

Methods

`VoiceModelFile`

Methods

`Synthesizer`

Methods

`AudioFeature` and `AudioFeatureSlice`

`VoicevoxCore`

Methods

`UserDict`

Methods

Enum: `AccelerationMode`

Enum: `OnExistingVoiceModelId`

Enum: `UserDictWordType`

`VoicevoxException`

​Onnxruntime

​Methods

​Constants

​OpenJtalk

​Methods

​VoiceModelFile

​Methods

​Synthesizer

​Methods

​AudioFeature and AudioFeatureSlice

​VoicevoxCore

​Methods

​UserDict

​Methods

​Enum: AccelerationMode

​Enum: OnExistingVoiceModelId

​Enum: UserDictWordType

​VoicevoxException

`Onnxruntime`

Methods

Constants

`OpenJtalk`

Methods

`VoiceModelFile`

Methods

`Synthesizer`

Methods

`AudioFeature` and `AudioFeatureSlice`

`VoicevoxCore`

Methods

`UserDict`

Methods

Enum: `AccelerationMode`

Enum: `OnExistingVoiceModelId`

Enum: `UserDictWordType`

`VoicevoxException`