Module hub 

The Hugging Face Hub calls the speaking model’s install makes: the repository’s current revision, its blobs with their LFS digests, and the download of one pinned file.

The token is carried for a gated repository; the speaking model’s own repository is open, so an absent token changes nothing.

Re-exports

use std::io::Write; 🔒

use std::path::Path; 🔒

use std::time::Duration; 🔒

use serde::Deserialize; 🔒

use serde::Deserialize; 🔒

use sha2::Digest; 🔒

use sha2::Sha256; 🔒

Structs

Hub

A Hub client with the optional token every request carries.

HubBlob

One file the repository publishes.

Lfs 🔒

ModelInfo 🔒

SpeakingFiles

The files one speaking variant is made of.

TreeEntry 🔒

Constants

CONNECT_TIMEOUT 🔒

The Hub’s connect and read bounds, a whole-request backstop sized so the largest variant (325 MB at 0.4 Mbps) still finishes, and the page cap on one listing before the install gives up.

MAX_TREE_PAGES 🔒

READ_TIMEOUT 🔒

TOTAL_TIMEOUT 🔒

Functions

finish_file

Check a finished download and move it into place, or remove it.

next_page

The next page a Link header names, when it names one.

parse_blobs

The files a tree listing carries, directories dropped.

part_path 🔒

Where a download is written while it is being checked.

speaking_files

The model file and the style files one variant is made of.

tree_url

The listing URL for one revision: recursive, because the model and its voices sit under onnx/ and voices/.

verify_release

Refuse a listing whose model file disagrees with the compiled variant.

voice_id

The voice id a style file’s path names. Every named voice carries a <language><gender>_<name> stem; the repository’s voices/af.bin blend does not, and is not one of the model’s named voices.