Aspeak

For CLI users

There are no breaking changes. But there are some differences.

Changes

- Now the CLI uses the REST API instead of the WebSocket API by default.
- You can use the `--mode websocket` flag to use the WebSocket API.
- You can also set the `mode` field to `websocket` in the auth section in your profile to use the WebSocket API by default.

Bug Fixes

- When the TTS API returns empty audio, aspeak no longer reports an cryptic "Unrecognized format" error.
- It will now report a warning in this case: "Got empty audio buffer, nothing to play"
- Now the voice listing command no longer fails with this API endpoint: https://speech.platform.bing.com/consumer/speech/synthesize/readaloud/voices/list

For Rust crate users

There are lots of breaking changes.

- Some fields of the `Voice` struct is now optional.
- We now uses an [modular approach for error handling](https://sabrinajewson.org/blog/errors) instead of using a big enum for all errors.
- Now there are two synthesizers: `RestSynthesizer` for the REST API and `WebSocketSynthesizer` for the WebSocket API.
- There is a `UnifiedSynthesizer` trait that provides a unified interface for both synthesizers.
- Some methods are renamed. For example, `Synthesizer::connect` is now `Synthesizer::connect_websocket`.
- Three new features of this crate:
- `rest-synthesizer`: Enable the `RestSynthesizer` struct.
- `websocket-synthesizer`: Enable the `WebSocketSynthesizer` struct.
- `unified-synthesizer`: Enable the `UnifiedSynthesizer` trait.
- The above three features are enabled by default.
- Disabling `websocket-synthesizer` feature will save you a bunch of dependencies. (`aspeak.rlib` is ~0.8MB smaller in release mode)
- Other minor changes.

For Python binding users

One breaking change:

- The `SpeechService` now automatically connect when it is constructed. The `connect` method is removed.
- It now uses the REST API by default.
- New keyword argument for the constructor of `SpeechService`:
- mode: `rest` or `websocket`. Default is `rest`.

CLI

You can now set the authentication secrets via the following environment variables:

- `ASPEAK_AUTH_KEY` for authentication using subscription key
- `ASPEAK_AUTH_TOKEN` for authentication using authorization token

Rust API

- Now you can use `Voice::request_available_voices`(or `Voice::request_available_voices_with_additional_headers`) to get the list of available voices.

- Add binary feature to aspeak crate to make rust lib less bloated
- From now on, building the CLI requires `-F binary` flag.

- Add binary feature to make rust lib less bloated

Enhancements

- Add support for `--color={auto,always,never}` options. And `aspeak` will also respect the `NO_COLOR` environment variable.
- There is an edge case that `aspeak` will use colored output even if `--color=never` is specified.
This is because `aspeak` uses `clap` to parse command line options. `--color=never` works only if the command line parsing is successful.
So if you specify an invalid option, `aspeak` will print the error message and exit. In this case, `aspeak` will use colored output.
- More documentation for the rust crate.
- Minor performance improvements.
- Now you can specify the custom voice list API url in your profile(field `voice_list_api` in section `auth`).

Breaking changes

- The default trial endpoint has been removed because it was shutdown by Microsoft. Now you must set up authentication to use `aspeak`.
- The default voice list API url has been removed for the same reason.
- The rust API has been changed.
- `Synthesizer` is now `Send`. Its various `synthesize_*` methods now takes `&mut self` instead of `&self`.
- Now you need to use the builder pattern to create various options like `TextOptions`.
- Fields of the `Voice` struct are now private. You can use the methods to access them.

Other changes

- The PKGBUILDs for Arch Linux is no longer stored in this repository. You can find them in the [AUR](https://aur.archlinux.org/packages/aspeak).