Added YouTube extraction to enable lyrics extraction. Added file upload section for addition of relevant info. #10
Conversation
| # Try to get the transcript (subtitles) in English | ||
| try: | ||
| transcript_list = YouTubeTranscriptApi.list_transcripts(video_id) | ||
| transcript = transcript_list.find_transcript(['en']) |
There was a problem hiding this comment.
One thing is that the song chosen by the user is very likely to be in their native language, so they would probably want to see the lyrics in their own language.
If we use subtitles, the user might have to select the language manually. That's why I was thinking of just using Whisper instead.
There was a problem hiding this comment.
Another thing is, if you extract YouTube using subtitles and your MP3 extraction uses Whisper, both methods achieve the same result but may behave differently, which could confuse the user.
There was a problem hiding this comment.
in regards to youtube subtitles and whisper I thought it would be nicer initially to have a failsafe so that if one fails the other kicks in. On review trying to use pure whisper I am encoutring lots of errors which now explains why i used the 2. Maybe, I may not know how but I thought the 2 would be nice to have them all.
There was a problem hiding this comment.
When I watch a Japanese video, I’d like to get the Japanese lyrics if I’m Japanese. If you want to provide this option, I think you can add another button to get the English subtitles. Let the user make the selection.
ai_generation_functions.py
Outdated
|
|
||
| # Read and return lyrics | ||
| with open("lyrics.txt", "r", encoding="utf-8") as f: | ||
| lyrics = f.read() |
There was a problem hiding this comment.
The result from Whisper is the lyrics, so why not use it directly instead of reading the temp file again?
ai_generation_functions.py
Outdated
|
|
||
| Your Thought: | ||
| {thought} | ||
| """ |
There was a problem hiding this comment.
User prompts have been moved to prompts.py. Please follow the rule.
|
|
||
| Please provide only the generated title in the specified language, without any additional explanation. | ||
| If the language is Chinese, use Traditional Chinese characters. |
There was a problem hiding this comment.
Traditional Chinese is not the same as the Simplified Chinese, so we need to specify it. (Or we need add more selection options at app.py)
prompts.py
Outdated
| Do not include any quotation marks or parentheses at the beginning or end of the title. | ||
| Your Thought: | ||
| {thought} | ||
| Please provide only the generated title in the specified language, without any additional explanation. Do not include any quotation marks or parentheses at the beginning or end of the title. |
There was a problem hiding this comment.
The change in this prompt doesn’t seem to have a very significant effect. What is the purpose of changing this prompt?
There was a problem hiding this comment.
The song title can be generated in the language we selected.
In this pull, I have taken into account the file addition requirements. Moreover polished the prompts.py as song titles were only being generated in Native Chinese. One main issue is that due to the youtube pipeline so far only enabling lyrics retrieval of songs in English its quite evident many languages are being hampered. Moreover, lyrics extraction is only so far for songs with captions enabled thus necessitating need for manual extraction to cater for all.