index

Effects

kokoro::download

effect kokoro::download {
  model: string;
  sizeBytes: number;
  source: string;
  dir: string
}

(source)

kokoro::speak

effect kokoro::speak {
  textLength: number;
  voice: string;
  outputFile: string;
  format: string
}

(source)

Functions

speak

speak(
  text: string,
  outputFile: string = "",
  voice: string = "af_heart",
  model: string = "fp32",
  speed: number = 1,
  allowedPaths: string[] = [],
  format: string = "",
  modelsDir: string | null = null,
): string raises <kokoro::download, kokoro::speak>

Speak text into an audio file. Returns the file path.

@param text - The text to speak, at most 50,000 characters @param outputFile - Where to write the file. Leave empty for a new temp file. An existing file is never overwritten. @param voice - A voice id from voices(), such as "af_heart" @param model - "fp32" or "q8" @param speed - Speaking speed, from 0.5 to 2 @param allowedPaths - Directories that outputFile must be inside @param format - "wav", "mp3", or "m4a". Leave empty to use the output file's extension, or wav when it has none. @param modelsDir - Where models are kept. Leave null for the default.

Text-to-speech that runs on this machine. After the model is downloaded once, no text is sent over the network. fp32 is faster. q8 is a smaller download: 92 MB instead of 326 MB.

The format comes from format when given, otherwise from the output file's extension, otherwise it is wav. An explicit format wins even when the extension says something else. mp3 and m4a are encoded by ffmpeg at 96 kbit/s, and speak refuses them before raising any interrupt when ffmpeg is missing.

Parameters:

Name	Type	Default
text	string
outputFile	string	""
voice	string	"af_heart"
model	string	"fp32"
speed	number	1
allowedPaths	string[]	[]
format	string	""
modelsDir	string | null	null

Returns: string

Throws: kokoro::download, kokoro::speak

(source)

download

download(
  model: string = "fp32",
  modelsDir: string | null = null,
): string raises <kokoro::download>

Download a Kokoro text-to-speech model. Returns the directory it is in. Does nothing when it is already there.

@param model - "fp32" (326 MB) or "q8" (92 MB) @param modelsDir - Where models are kept. Leave null for the default.

Fetches a model before the first speak call needs it. Calling this is optional: speak downloads a missing model itself. The directory is chosen the same way speak chooses it.

Parameters:

Name	Type	Default
model	string	"fp32"
modelsDir	string | null	null

Returns: string

Throws: kokoro::download

(source)

voices

voices(): Voice[]

List the Kokoro voices. Each voice has an id, name, language, gender, and a quality grade from A (best) to F.

Returns: Voice[]

(source)