The following are experiments and examples created to explore approaches to the presentation of information derived from the proceedings of Old Bailey Trials (Sessions Papers). The dataset that this work relies on is output of 'The Old Bailey Online' and related projects.
The contents of this page are likely to change regularly, please get in touch if you want to know more or want stable copies.
Ben -> ben.jackson@sussex.ac.uk
Old Versions and Tests

Links in this Column are old version - These are only preserved for the benefit of recording progress - The current version should contain the accumulation of this work.

Old Bailey - speech synthesis tests - part 1 (X3DOM)

I hear voices - 1 - Textured (Ubuntu Chrome)

I hear voices - 2 - No Textures (Ubuntu Chrome)

I hear voices - 3 - Textured (mac Chrome)

I hear voices - 4 - No Textures (mac Chrome)

I hear singing - 5 - Textured (mac Chrome)

I hear voices - 6 - Textured (windows Chrome)


Speech Synthesis tests (1) – Experimenting with the W3C Web Speech API using commonly available speech synthesisers provided by web browsers. The target application is a large set of utterances/speech-acts taken from historical sources. Text to speech could provide a convenient and low resource solution for spoken representation. Use of common web browser technology provides widespread availability and ongoing improvements to the technology will improve quality over time. Future work may explore alternative systems and recordings of human actors to attempt improved sonic quality and listener experience. The current design choices favour avoiding attempted realism to reduce the misrepresentations this can create.

The outcome of these tests was the decision to use the speech synthesis features provided by Google Chrome, Chrome voices reproduced the words and abbreviations present in the sample data with greater (subjective) success than alternative synthesisers. There was no standard set of voices available during testing, operating systems and browsers varied considerably - Which means it is currently necessary to add selection logic to handle the variations.

A problem associated with using text to speech in web browsers is the number of voices available, in one case this meant only 1 male and one female voice was available, as a result manipulation of pitch and speed variables was attempted to distinguished between the characters in the trial.

A significant issue affecting the use of Google voices is that working with them required non-standard code, without these workarounds the synthesiser become unresponsive and blocked when reproducing long passages. The workarounds interfere with the operation of standard compliant synthesisers, so additional selection logic is required to overcome the differences. Fixing this is not a priority at the moment, the issue appears to be a bug/error in the Chrome voices speech synthesis engine.

The singing versions will only work if your system provides the selected voices, in my case these were only available on my Apple computer, my interest in the singing voices is that the performance reproduce the words at a slower speed than the spoken ones, which may affect consumption of the story. Firefox on a Linux machine used during this section of development included a selection of regional UK dialects.