https://github.com/nacnudus/psalter
Psalter web app maybe one day
https://github.com/nacnudus/psalter
Last synced: 2 months ago
JSON representation
Psalter web app maybe one day
- Host: GitHub
- URL: https://github.com/nacnudus/psalter
- Owner: nacnudus
- Created: 2024-08-08T15:02:59.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-10-27T20:40:56.000Z (6 months ago)
- Last Synced: 2025-01-05T18:26:48.485Z (4 months ago)
- Language: Python
- Size: 797 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Psalter
One day this might be a smartphone app similar to the website
https://www.rmjs.co.uk/psalter/psalter.php.
So far it's merely a JSON representation of the psalms.
## Wordlist
https://dhanswers.ach.org/topic/creating-a-wordlist-from-text/#post-1762
```sh
tr -sc '[A-Z][a-z]' '[\012*]' < psalmtext.txt | sort | uniq > wordlist.txt
```
## Transcription and alignment
Review https://github.com/nacnudus/audio-verse-splitter
## Consolidate timestamps files
```sh
find audio/priory-1 -type f -name "timestamps.txt" | xargs -I {} bash -c 'cp "$1" temp/$(basename $(dirname "$1")).txt' -- {}
```
## Copy a directory structure without files
```sh
rsync -a --include '*/' --exclude '*' recordings/priory-1 "timestamps"
```
## Split audio files at timestamps
```sh
find recordings/st-pauls-scott -type f -name "*.flac" | xargs -I {} bash -c './split.sh $1' -- {}
```
## Calculate the actual durations of audio files
```sh
fd -I --exec sox {} -n stat 2>&1 | grep "Length" | awk '{print $3}' > durations.txt
```
## Create audio samples and metadata
Use script `manifest2metadata.R` to create a JSON manifest of each audio file
and its transcription.
Also concatenate parts of a verse into a single file, as long as the total
length is within the limits of the Whisper model, which is 30 seconds. Doing so
is more efficient, because it will halve the amount of samples to use in
training, and it mitigates the inaccuracy of splitting parts of verses that,
when sung, did not pause at the colon.
## Create a Huggingface dataset
https://me.sakana.moe/2024/09/03/a-complete-guide-to-fine-tuning-and-deploying-whisper-models/
If huggingface login and dataset upload hang, you might have to disable IPv6.
```sh
sudo sysctl -w net.ipv6.conf.all.disable_ipv6=1
```