Task list, 28/9/2021

Today

• separation script: spleeter: see run_spleeter.py

• Extend abair xml to return list of timestamps; segment long recordings: notebook

• Rebase w2v notebook on this or this

• Add LM and timings: see here, repo, file, this issue, parlance/ctcdecode, wav2vec2_kenlm.py

• Fingerprint for known audio: dejavu

• Pass over input data, with this or something similar

• MFA, based on this

Look into:

• Add official ASR CTC example to examples/pytorch/speech-recognition
• Rewrite padding logic from pure python to numpy
• Non-Adversarial Unsupervised Word Translation
• Phonetic-and-Semantic Embedding of Spoken Words with Applications in Spoken Content Retrieval
• grtzsohalf/Audio-Phonetic-and-Semantic-Embedding
• SpeechToolsWorkers

Personal

• Run this See this:

--match-filter "license='Creative Commons Attribution license (reuse allowed)'"

Living audio

Longer term

• TG4 Foghlaim scraper Lessons

• Scrape more Ros na Rún

• Compare this with stuff from last year

• Segmentation: run_cleanup_segmentation.sh, tedlium, AMI

• VOSK LM

• CUNY-CL

Look at:

• 2dot71mily/youtube_captions_corrections
• microsoft/Recognizers-Text
• hiromis/notes
• Alexander-H-Liu/NPC
• andi611/Mockingjay-Speech-Representation
• jina-ai/jina
• Continue this — p. 18
• scrapinghub/portia
• wav2vec2-large-voxrex, Kungbib/swedish-bert-models