Open maboa opened 1 year ago
For context, Gentle is a forced aligner that (provided an audio file and a transcript) will generate per-word (and even sub-word) timecodes. Their source code is available here on Github.
🔰 I'm attaching two example Gentle JSON files (for the same recording) below. ASR Timed Text Format Test 2 [Gentle] P.json
This second JSON file uses homophone substitutions for words that are not in Gentle's pronunciation dictionary in order to acquire more reliable phoneme timecodes for more words. ASR Timed Text Format Test 2 [Gentle] H.json
The transcripts have some minor custom markup. [+] = the beginning of a sentence \\ = the end of a sentence || = the end of a paragraph
The corresponding audio file can be obtained here.
Also note that although HyperAudio Converter is able to import Gentle files and convert them to HyperAudio hypertranscripts, the formatting of the HyperAudio hypertranscript is outdated and cannot be directly imported by HyperAudio Lite Editor.
@natelawrence I would love it if you added read/write support for Gentle alignments and YouTube Automatic Captions XML/VTT formats that contain per-word timestamps.
I'd like to put a $150 bounty on both: Hyperaudio editor reads/writes Gentle JSON. Hyperaudio editor reads/writes YT AC XML.
https://twitter.com/natelawrence/status/1653373649758605313?s=20