Matxa Multiaccent Endpoint

Input parameters

Parameter	Description	Type	Options/Range	Recommended Values
text	Input text to be converted to speech	string	Any text input	N/A
voice	Speaker name for voice output	string	8 speakers available	N/A
accent	Selected dialect for voice output	string	balear, central, nord-occidental, valencia	N/A
temperature	Controls sampling variance during inference. Lower values yield higher quality but less variability	float	0.2 to 0.67	0.2 to 0.67
length_scale	Related to speech speed. Higher values make the speech slower, while lower values make it faster	float	0.75 to 0.9	0.75 to 0.9

List of associated accents and speakers (female voice and male voice) available. Each integer value is associated with one of the speakers you see in the list below:

Accent	Name	speaker_id
Balear	Quim	0
Balear	Olga	1
Central	Grau	2
Central	Elia	3
Nord-occidental	Pere	4
Nord-occidental	Emma	5
Valencia	Lluc	6
Valencia	Gina	7

Code examples

Python

example.py

import requests

API_URL = "https://p1b28cv1e843tih1.eu-west-1.aws.endpoints.huggingface.cloud/api/tts"
headers = {
   "Authorization": "Bearer <hf_token>",
}

def query(text):
   data = {"text": text, "voice": "quim", "accent": "balear"}
   return requests.post(API_URL, headers=headers, json=data)

response = query("Bon dia")
with open("output.wav", "wb") as f:
   f.write(response.content)

Curl

One effective way is to create a temporary JSON archive to pass text and input parameters for inference.

printf '%s' '{
  "text": "Ara caldrà veure què passa amb aquestes investigacions judicials, si segueixen endavant i se l'\''acaba jutjant, o si es desestimen d'\''acord amb la immunitat que tindrà un altre cop com a president del país.",
  "voice": “quim”,
  “accent”: “balear”,
  "type": "text"
}' > data.json

curl -X POST https://p1b28cv1e843tih1.eu-west-1.aws.endpoints.huggingface.cloud/api/tts -H "Content-Type: application/json" -H "Authorization: Bearer <hf_token>"  -d @data.json | play -t wav -

Javascript (Nodejs)

Executed with Nodejs. Install NPM (Node Package Manager) and with NPM install fetch-node library.

const fetch = require("node-fetch");
const fs = require("fs");

// Define the API URL and headers
const API_URL = "https://p1b28cv1e843tih1.eu-west-1.aws.endpoints.huggingface.cloud/api/tts";
const headers = {
    "Authorization": "Bearer <hf_token>",
    "Content-Type": "application/json"
};

// Function to send the request
async function query(text) {
    const data = {
        text: text,
        accent: “balear”,
        voice: “olga”
    };

    try {
        // POST request
        const response = await fetch(API_URL, {
            method: "POST",
            headers: headers,
            body: JSON.stringify(data)
        });

        // Check the response
        if (!response.ok) {
            throw new Error(`Error: ${response.status} ${response.statusText}`);
        }

        // Convert the response to a buffer
        const buffer = await response.buffer();

        // Write the buffer to an output file
        fs.writeFile("output.wav", buffer, (err) => {
            if (err) {
                console.error("Error saving the file:", err);
            } else {
                console.log("File saved as output.wav");
            }
        });

        
    } catch (error) {
        console.error("Error making request:", error);
    }
}

// Your query
query("Bon dia.");