Spaces:
Paused
Paused
import { getInference } from '$lib/agents/getInference'; | |
import type { TextToSpeechArgs, TextToSpeechOutput } from '@huggingface/inference'; | |
import type { Tool } from './tool'; | |
export const textToSpeechTool: Tool<TextToSpeechArgs['inputs'], TextToSpeechOutput> = { | |
name: 'textToSpeech', | |
description: 'This tool takes a text input and turns it into an audio file.', | |
examples: [ | |
{ | |
prompt: 'Say the following out loud:"Hello world!"', | |
command: "textToSpeech('Hello world!')" | |
}, | |
{ prompt: 'Say the content of the string txt out loud', command: 'textToSpeech(txt)' } | |
], | |
call: async (input) => { | |
return await getInference().textToSpeech( | |
{ | |
inputs: await input, | |
model: 'espnet/kan-bayashi_ljspeech_vits' | |
}, | |
{ wait_for_model: true } | |
); | |
} | |
}; | |