How to Use ACE-step as a Python Library?

[Yangukon]

I’m struggling and can’t figure out how to use ace-step as a Python library with its API.
First, I run pip install git+https://github.com/ace-step/ACE-Step.git, and then… I have no idea what to do next.
There are no examples, just me and the void of ignorance.
Any help or example would be greatly appreciated!

[allo-]

Does this example help?

https://github.com/ace-step/ACE-Step/blob/main/infer-api.py

[Dimasiggs]

I tried with this example, but I get an error.
Response: {'detail': "Error generating audio: object of type 'int' has no len()"}

as I understand it, the parameters are somehow not transmitted correctly and because of this, infer_step was transmitted instead of text

  File "ace-step-test\venv\Lib\site-packages\acestep\pipeline_ace_step.py", line 1519, in __call__
    if len(lyrics) > 0:
       ~~~^^^^^^^^
TypeError: object of type 'int' has no len()

here is the code i used to send the request:

import requests
import json

url = "http://localhost:8000/generate"
headers = {
    "Content-Type": "application/json"
}

data = {
    "prompt": "upbeat pop song",
    "duration": 10.0,
    "steps": 27,
    "guidance_scale": 7.5,
    "checkpoint_path": "checkpoint",
    "audio_duration": 10.0,
    "lyrics": "lol",
    "infer_step": 27,
    "scheduler_type": "ddim",
    "cfg_type": "standard",
    "omega_scale": 1.0,
    "actual_seeds": ["123456"],
    "guidance_interval": 1,
    "guidance_interval_decay": 0.0,
    "min_guidance_scale": 1.0,
    "use_erg_tag": False,
    "use_erg_lyric": False,
    "use_erg_diffusion": True,
    "oss_steps": [10]
}

try:

    response = requests.post(url, headers=headers, data=json.dumps(data), timeout=30)
    print(f"Status code: {response.status_code}")
    print("Response:", response.json())
except requests.exceptions.RequestException as e:
    print("Request error:", e)

[allo-]

I think it is a bug in the infer-api script introduced by the new format parameter.

Here it uses *args using duration as first parameter:

ACE-Step/infer-api.py

Line 62 in d612523

input_data.audio_duration,

And here is a first parameter for the format:

ACE-Step/acestep/pipeline_ace_step.py

Line 1434 in d612523

format: str = "wav",

If you really want to infer from python, I'd copy the code from the infer script instead of using the HTTP API and then use **kwargs setting (only) the parameters you need. The infer-api script would need a parameter input_data.audio_duration="wav" in line 62.

I can later test myself and do a pull request, or you can do if you get it fixed by that.

[allo-]

Try to use #279 and this version of your client script:

import requests
import json

url = "http://localhost:8000/generate"
headers = {
    "Content-Type": "application/json"
}

data = {
    "format": "mp3",
    "prompt": "upbeat pop song",
    "duration": 10.0,
    "steps": 27,
    "guidance_scale": 7.5,
    "checkpoint_path": "checkpoint",
    "output_path": "output",
    "audio_duration": 10.0,
    "lyrics": "lol",
    "infer_step": 27,
    "scheduler_type": "euler",
    "cfg_type": "apg",
    "omega_scale": 1.0,
    "actual_seeds": ["123456"],
    "guidance_interval": 1,
    "guidance_interval_decay": 0.0,
    "min_guidance_scale": 1.0,
    "use_erg_tag": False,
    "use_erg_lyric": False,
    "use_erg_diffusion": True,
    "oss_steps": [],
    "output_path": "test.mp3"
}

try:

    response = requests.post(url, headers=headers, data=json.dumps(data), timeout=30)
    print(f"Status code: {response.status_code}")
    print("Response:", response.json()["detail"])
except requests.exceptions.RequestException as e:
    print("Request error:", e)