The following notes, compiled by Eugene Yue, give details of the speech synthesizers included in the first of the recordings below. The second recording is of Aeneid IV: 1-53.
Poeta ex Machina (https://poetaexmachina.net/) works online but the audio can be downloaded. It uses classical pronunciation. Apart from no "h" and incorrect "y", it does not differentiate "ph" from "p", etc. It cannot read macronised text but does its own macronisation. Ambiguous vowel quantity is resolved with metre and is not accurate for prose. Users cannot modify the speech synthesis. It reads the above text as:
eSpeak (http://espeak.sourceforge.net/) is a TTS programme. The programme and its engines require installation. The engines are compatible with most other TTS programmes requiring installation. There is no problem saving the audio or modifying the speech synthesis. Its Latin engine reads macronised text with correct classical pronunciation and there is no Ancient Greek engine other than that of eSpeak. However, the sound is too robotic.
Google Translate (https://translate.google.com/) works online and the audio cannot be downloaded. Its Latin engine reads text, macronised or not, with defective Italianate pronunciation and no differentiation between long and short vowels.
meSpeak.js (https://www.masswerk.at/mespeak/) is an online port of eSpeak and the audio cannot be downloaded. It cannot read macronised text and users cannot modify the speech synthesis.
ResponsiveVoice (https://responsivevoice.org/text-to-speech-languages/unde-orationem-text-in-latin/) works online. The audio cannot be downloaded and users cannot modify the speech synthesis. Using the Google Translate Latin engine, it reads text, macronised or not, with defective Italianate pronunciation and no differentiation between long and short vowels.
II Non-Latin engines (with converted text):
Among all Microsoft and compatible non-Latin/Ancient Greek engines, Microsoft Cantonese engines are the only ones having the required sounds to read Latin text (macronised and converted) with classical pronunciation and Ancient Greek text (converted) with reconstructed/school pronunciation satisfactorily. On the other hand, most of the Italian engines can read Latin text (macronised and converted) with correct Italianate pronunciation as well as differentiation between long and short vowels. There is no problem saving the audio or modifying the speech synthesis.
Install the following: 1. Balabolka 2. Microsoft Text-to-Speech Languages for Windows 10 (Hong Kong for Latin with classical pronunciation and Ancient Greek; Italian for Latin with Italianate pronunciation; follow the answer to the second question under Speech Engines in the Balabolka FAQ to unlock the engines)
Save the following files in the Balabolka folder in the Documents folder of the computer, launch Balabolka and select the SAPI 5 platform, speech engine and appropriate saved files (to appear in the Panel of Dictionaries): Latin with classical pronunciation 1 Latin - Classical 1.rex 1 Latin - Classical 2.rex Latin with Italianate pronunciation 1 Latin - Italianate 1.rex 1 Latin - Italianate 2.rex Ancient Greek with reconstructed pronunciation 2 Ancient Greek - Reconstructed.rex Ancient Greek with school pronunciation 2 Ancient Greek - School.rex Optional 1 Latin - i to j.bxd (for text properly macronised and without j) 1 Latin - i to j & u to v.bxd (for text properly macronised, without j and with u for v) 1 Latin.bxd (for text without macrons; not very good given many homographs) 1 Latin - Italianate 1a.rex (for pronunciation without differentiation between long and short vowels)
Input text (for Latin, properly macronised with j; otherwise, select one of the three relevant optional files) and play/save the audio.
Earlier conversion and instructions files for application of the approach with Word are also attached. They are more difficult to use and require a number of fixes. However, they show the enormous flexibility of the approach for adaptation and provide hints on customisation of pronunciation