This document provides a short manual for Bellbird.

Bellbird is written in C, and is intended to run on Linux or win32 systems. It does not support other systems since no test hardware is available to the current developers. If developers wish to join, other systems could be supported.

Here are some permutations of usage examples.

./bellbird --voice voice.flitevox -f inputtext -o play

This will read inputtext and play out to ALSA using the Clustergen voice (voice.flitevox)

./bellbird --voice cmu_us_arctic_slt.htsvoice -f - -o outputfile.wav

This will read STDIN and output to a wav file using a hts voice (cmu_us_arctic_slt.htsvoice in the current working directory)

./bellbird --voice voice.flitevox -f inputtext -o - | opusenc --bitrate 128 - outputfile.opus

This will read inputtext and send it to STDOUT where it can be grabbed by opusenc (if that is your preferred and installed encoder).

./bellbird --voice cmu_us_arctic_slt.htsvoice -f inputtext -o - | lame -b 128 - outputfile.mp3

This will read inputtext and send it to STDOUT where it can be grabbed by lame (if that is your preferred and installed encoder).

Bellbird is open source software. It contains code from a number of projects including flite, hts_engine, festival and flite+hts. Please see the COPYING file for information on the licenses. Note all the licenses are broadly similar to the original Flite license. Bellbird uses the same license to allow the exchange of code. Bellbird has benefitted extraordinarily from the ability to use other projects code for creating Bellbird.

Bellbird was branched from Flite for Android's version of: Flite: a small run-time speech synthesis engine version 1.5.6-current

It contains code from:

• flite 2.0 (Dec 2014)
• hts_engine 1.08 (Dec 2013)
• flite+hts 1.04 (Dec 2012)

Bellbird contains material from significant numbers of contributors to a number of different projects. The ACKNOWLEDGEMENTS file attempts to collate these contributions. Very little of Bellbird was written by Bellbird's authors. Bellbird's authors wish to thank the other projects whose code made Bellbird possible. Bellbird's authors wish the best for the these other projects and look forward to their continued vitality. Any deficiencies in Bellbird are solely the responsibility of Bellbird's authors.

Bellbird is quick and easy to build with minimal dependencies. Build instructions may be found in doc/building.md

Bellbird voices are all dynamically loaded at Bellbird startup. Bellbird support two voices types at the current time.

1. The argument "--voice" should be set to a full pathname of dumped Clustergen voice.
2. The argument "--voice" should be set to a full pathname of HTS (version 1.07) voice.

Voice files for testing purposes are collected at https://github.com/peterdrysdale/bellbird_extras.

Note voices are a few megabytes in size but there will still be some time required to read them into newly created structures. On my 700Mhz netbook there is a 0.2 sec overhead to read in the voices relative to mapped Flite voices. Bellbird is not targetted for embedded work. Bellbird recommends the Flite library for such applications.

Bellbird is not intended as a fork of flite, flite+hts or festival, in the sense that we do NOT wish to compete with these worthwhile projects. This is the reason we have labelled it a remix. In music remixes are rarely intended be a competitor to the orginal work.

Bellbird was originally a toy of its originator while he acted as an uploader for maintaining the Debian package of festival. It was used originally to test ideas and his understanding of festival to allow him to maintain the Debian packaging for festival. Additional features have been added to Bellbird.

Bellbird offers the user the following preferred features.

1. Dynamic (only) loading of a range of voice types including clustergen (versions v1.5.6 and v2.0 flitevox) and hts voices.
2. STDIN and STDOUT, input and output redirection via commandline. This is designed to allow flexible preprocessing of input text and allow interfacing to a wide range of audio conversion software for generated sound data. It is easy to send Bellbird sound data to opus, mp3 and other encoders for further processing.
3. UTF-8 directional apostrophe support for contraction type words. With the rise of the web, UTF-8 punctuation in English text is more common. Bellbird will accept UTF-8 apostrophes in addition to ASCII formatted text. Flite v2.0 has now added this functionality also.
4. UTF-8 quotation mark support.
5. Allow UTF-8 symbols in dictionary words. It is now common in English for loan words to be written with UTF-8 accent symbols. Bellbird allows such words to be retrofitted to the ASCII dictionary.(Note automatic case conversion doesn't occur for UTF-8 symbols).
6. Bellbird is able to pronounce numerous U.K./Commonwealth spelling variants of English words for U.S. voices (these are generally pronounced with U.S. accent). The previous cmu lexicon was confined to U.S. English spelling. A user (even if in the U.S.) has no control on the spelling variant of text available either in classic novels or on the internet. It is preferrable that TTS will read this in the same way that a U.S. reader will adapt to variant spelling. (Often such readers will not even notice which variant they are reading.) Words are added as they are heard by the authors during the operation of Bellbird. Bellbird does not have a complete U.K. spelling variant pronouncing dictionary at this time but it already has many common and not so common U.K. spelling variants. The authors are now rarely hearing U.K. spelling variants whose pronounciation is not upto the same standard as U.S. spelling variants when pronounced by the U.S. voices.
7. Bellbird is actively improving its pronouncing dictionary as use of Bellbird highlights words which could have improved pronounciation.
8. Large File support. The WAV format allows a maximum file size of 4Gb. Where the O.S. supports this, Bellbird will write WAV files in excess of 2Gb upto the format maximum of 4Gb. Users who wish to write more than 4Gb of sound data are recommended to use Bellbird's STDOUT redirection and use audio software such as an opus or mp3 encoder to reformat the data.

Bellbird offers the developer, the following features:

1. compiles clean using gcc and clang with -Wall and -Wextra,
2. regular testing and compliance against the latest compilers (Developers test against the latest gcc or clang as they become available in Arch or Debian testing),
3. Autotools or cmake build environment with incremental builds,
4. minor performance improvements,
5. rapid full tree rebuilds.

In the long tradition of open source, Bellbird "scratched the itch" of one person. It is being developed publicly in case others might find it useful and/or other programmers wish to borrow code from it.

Bellbird welcomes other developers to assist in developing Bellbird as a community based open source package. (Although we don't yet support BSD, we would welcome our friends from the BSD communities if they wish to join). Any code which would better benefit festival, flite or hts_engine should be sent to those projects first and in preference.

How much Bellbird develops depends on if other people find it useful and provide feedback to Bellbird.

Bellbird at this time is not yet expected to be embedded into other applications. Bellbird recommends programmers consider the Flite library for such purposes.