Add YuE (music gen) from fal.ai

[Wauplin]

Similar to huggingface/huggingface.js#1148.

Will keep it as draft for now. I mostly wanted to see how that integration would work given that it doesn't use the same parameters as we have for text_to_speech (has an extra "genres" parameters). For now I added it to the signature + docstring but it's not compliant with https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/src/tasks/text-to-speech/spec/input.json.

Another possibility is to add a text-to-audio or text-to-music but that seems premature given it's a single model, single provider and still experimental. Will have to wait more example before an harmonization.

Working code example:

import os
from huggingface_hub import InferenceClient

lyrics = """
[verse]
In the town where I was born
Lived a man who sailed to sea
And he told us of his life
In the land of submarines
So we sailed on to the sun
'Til we found a sea of green
And we lived beneath the waves
In our yellow submarine

[chorus]
We all live in a yellow submarine
Yellow submarine, yellow submarine
We all live in a yellow submarine
Yellow submarine, yellow submarine
"""
genres = "pavarotti-style tenor voice"

t0 = time.time()
client = InferenceClient(
    provider="fal-ai",
    model="m-a-p/YuE-s1-7B-anneal-en-cot",
    api_key=os.getenv("FAL_AI_KEY"),
)
audio = client.text_to_speech(lyrics, extra_parameters={"genres": genres})
with open("output.mp3", "wb") as f:
    f.write(audio)

DnwXm1UGLGJZGPPFGbnL.mp4

(what a masterpiece, I know)

Process took 594s (~10min) to run on fal.ai. This means it will fail if we go through HF proxy as there is a 60s timeout. For now only works when passing a FAL_AI key directly. This can be fixed using async/queuing mechanism but out of scope for this PR.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[hanouticelina]

maybe genres can be passed inside extra_parameters when #2812 is merged.

[Wauplin]

Waiting for it! 😄

[hanouticelina]

here you go : 07e1adb 😄

[Wauplin]

I wonder in the future how we will distinguish between "lyrics" (for music-gen) and "text" (for text-to-speech). For now there's only 1 model so let's do it this way.

[hanouticelina]

yes, same issue with replicate for this task 😕

[hanouticelina]

thank you!