@faro@sigmoid.social Audio-AI researcher at @audioshakeai (Before: @inria, @FraunhoferIIS / @uniFAU). All in 17.68% of grey
Montpellier, France
Joined April 2008
- Tweets10K
- Following715
- Followers1K
- Likes845
FaRo retweeted
DJs making music magic, create yours with AudioShake stems. djay Pro from @algoriddim now includes AudioShake’s tech to isolate vocals, instruments, and drums at the highest quality available to the industry. On mobile. 🎧📷#djtools #beatmaking
FaRo retweeted
Our paper on lyrics transcription evaluation is on arXiv with updated and extended results (including Whisper v3)!
📄 arxiv.org/abs/2311.13987
Also, the benchmark is now on @paperswithcode:
🏆 paperswithcode.com/dataset/j…
👏 @faroit @h_schreiber @Luke_Miner @cnst_ant @AudioShakeAI
This month at #ISMIR2023, AudioShake’s Research team presented a new benchmark for automatic lyric transcription systems– one that accounts for the nuances of music. You can read more on their new paper on AudioShake: audioshake.ai/post/new-bench…
FaRo retweeted
We just released Jam-ALT, a formatting- and punctuation-aware automatic lyrics transcription benchmark (based on the JamendoLyrics dataset) that follows music industry guidelines. 🧵
🔎 audioshake.github.io/jam-alt…
🤗 huggingface.co/datasets/audi…
🧑💻 github.com/audioshake/alt-ev…
FaRo retweeted
#Pytorch implementation of MusicLM, new SOTA model for music generation using attention networks plus embeddings from MuLan, a text-audio contrastive learned model github.com/lucidrains/musicl…
FaRo retweeted
🎻 The SDX23 challenge introduces a new formulation of audio source separation: cinematic sound separation.
The task is to separate a movie's audio into three tracks: dialogue, sound effects & music.
📕 Give it a try using the starter kit.
aicrowd.com/challenges/sound…
FaRo retweeted
#AudioLDM, the text-to-audio model, is now available on HuggingFace and GitHub to play with!
We will add more functionality and further improve the model performance in the near future.
Share the interesting samples you generate!
github.com/haoheliu/AudioLDM
huggingface.co/spaces/haohel…
FaRo retweeted
Lots of insights for ML researchers in this great interview with @colinraffel by @ellieevsss @forai_ml
Some highlights:
We're in the dark ages of ML model development. How can we build models collaboratively like we build open-source software?
youtube.com/watch?v=fDUXSoFS…
-1-
FaRo retweeted
After 16 years, Twitterrific is discontinued: killed by an unannounced and undocumented policy change from a company no developer should ever trust again blog.iconfactory.com/2023/01…
FaRo retweeted
Diffusion Transformers (DiTs) are a new class of latent diffusion models (LDM) based on the transformers, obtained replacing the U-Net backbone with a transformer that operates on latent patches. DiTs are very similar to Vision Transformer (ViT)
github.com/facebookresearch/…
FaRo retweeted
Chatbots are not a good replacement for search engines
iai.tv/articles/all-knowing-…
FaRo retweeted
A great read. Stop using the elbow criterion for k-means and how to choose the number of clusters instead (alternatives).
"..researchers and reviewers should reject conclusions drawn from the elbow method."
📄 Paper: arxiv.org/abs/2212.12189
FaRo retweeted
FlexiViT: One Model for All Patch Sizes
abs: arxiv.org/abs/2212.08013
github: github.com/google-research/b…
FaRo retweeted
OpenAI Whisper blew everyone's mind with its translation and transcription. But 1-thing was missing "Speaker Diarization"
Thanks to @dwarkesh_sp code, we have it right infront as a @Gradio app on @huggingface Spaces.
⚙️ - huggingface.co/spaces/dwarke…
🎥 - youtube.com/watch?v=MVW746z8…
FaRo retweeted
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
We propose a text-queriable sound separation model that can be learned w/o text-audio pair data.
Work is done with and first authored by @salu133445 during an internship @Sony.
arxiv.org/abs/2212.07065
