audio field — a nested object keyed by TTS model and voice, or null if no audio exists.
Response shape
bitrate is in kbps, duration in seconds, and fileSize in bytes.
Available models and voices
Coverage varies per paragraph. Only thetts-1-hd / nova combination covers every paragraph. Read the audio object rather than building URLs by hand — a voice that is absent for a paragraph is absent from the object.
Accessing audio
Audio URLs are included in responses from all paragraph-returning endpoints:/search, /paragraphs/*, /papers/:id, and the dedicated /audio/:ref endpoint.
cdn.urantia.dev via Cloudflare CDN. See CDN & Static Assets for the URL patterns and for paper-level audio.
Audio for a whole paper
GET /papers/{id} also returns paper.audio: one MP3 file for the whole paper, with the time where each item starts in that file. An item is the paper title, a section title, or a paragraph. Use it to play a paper from one file and to show which paragraph and sentence the voice reads.
keyistitle,section:{sectionId}, or the standard reference of a paragraph.atis the start of the item in the file, in seconds. The starts are in reading order and ascend.sentencesis the start of each sentence of a paragraph, in seconds. The first value is the same asat. A paragraph whose sentences have no times has nosentences.- An item ends where the next item starts. The last item ends at
duration. - Section 0 of a paper has no title, so it has no
section:0item. audioisnullfor a paper with no time map.
at. To find the paragraph that sounds, find the last start that is at or before the current time.
One file for a paper is good for a phone with the screen off: some browsers stop or delay a change from one file to the next while the screen is off. In one file, the voice continues with no change of file.
A speech model checked each start against the words of its item. Some starts were checked by ear.
