Sagen (German for "to say") is my attempt at making a text-to-speech engine aimed at .NET developers who don't have thousands of dollars at their disposal to license a commercial speech synthesis solution. In many ways, it is an experiment and continual learning experience for me, as I am not an expert in speech science, phonetics, or vocal acoustics; I simply want to see how far I can go with original research, freely available resources, and lots of patience.
There are tons of TTS engines out there already, but I feel like they're all missing something.
Aside from being often prohibitively expensive, it is not unusual for commercial TTS systems to be restrictive in their available customizations for voices, voice parameters, and context-sensitive vocal qualities (e.g. intonation, stress, and timbre). Such qualities are necessary to convey meaning in speech.
Concatenative synthesizers, as well as other similar "realistic" TTS technologies tend to be CPU-heavy, leave a large memory footprint, and require each voice to be installed separately. Because they are based on databases of recorded speech samples, they are not very customizable at all.
There are also many free options for speech synthesizers, but they often have sparse, confusing, or convoluted documentation, or are locked down to one specific language (e.g. Java). While all TTS libraries have advantages and disadvantages, I feel like the .NET crowd would welcome a TTS solution specifically made for them.
My goal with Sagen is not necessarily to produce "something better", but to instead offer a user-friendly TTS engine with a respectable amount of configurability, flexibility, and performance. The best part? It's free.
Here is a short list of major features that will be supported:
- Text-to-speech based on an articulatory model adapted from Voc
- Plentiful parameters for tuning how voices sound (age, sex, vocal force, hoarseness, etc...)
- Support for direct playback, WAV exporting, and sending audio data via
System.Stream - Multiple options for sample format and rate (export only)
- Support for X-SAMPA-based pronunciation lexicons
- Multiple language support (English and German are currently prioritized)
- Heteronym resolution
- Singing?!
It is currently a heavy work-in-progress, and I welcome your input and/or contributions.
This project is made available under the MIT License and is completely free for anyone to use, for any purpose, without the burdens of licensing costs or royalties.
This project contains code adapted from Paul Batchelor's Voc library, which is in turn based on Neil Thapen's bizarre but fascinating project, Pink Trombone.
See LICENSE.md for all licenses and copyright notices.