Deploy HARP wrapper via model agent

.harp/manifest.json

@@ -0,0 +1,30 @@
+{
+  "backend_space": "Soul-AILab/SoulX-Singer",
+  "deploy_mode": "remote-backend",
+  "entry": "app.py",
+  "framework": "gradio_client",
+  "generated": true,
+  "io": {
+    "inputs": [
+      "audio",
+      "audio",
+      "dropdown",
+      "checkbox",
+      "slider",
+      "number",
+      "dropdown",
+      "dropdown",
+      "checkbox",
+      "checkbox"
+    ],
+    "outputs": [
+      "audio",
+      "file",
+      "file"
+    ]
+  },
+  "repo_id": "Soul-AILab/SoulX-Singer",
+  "source": "recipe",
+  "space_layout": "huggingface-gradio",
+  "task": "text-to-audio"
+}

README.md

@@ -1,13 +1,19 @@
 ---
-title: SoulX Singer
-emoji: 🌍
-colorFrom: red
-colorTo: green
+title: "SoulX-Singer"
+colorFrom: indigo
+colorTo: gray
 sdk: gradio
-sdk_version: 6.19.0
-python_version: '3.13'
+sdk_version: 5.28.0
 app_file: app.py
 pinned: false
+license: "apache-2.0"
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# SoulX-Singer
+
+SoulX-Singer is a high-fidelity, zero-shot singing voice synthesis model that enables users to generate realistic singing voices for unseen singers. It supports melody-conditioned (F0 contour) and score-conditioned (MIDI notes) control for precise pitch, rhythm, and expression.
+
+- Inputs: audio, audio, dropdown, checkbox, slider, number, dropdown, dropdown, checkbox, checkbox
+- Outputs: audio, file, file
+
+Generated by the HARP model agent from a recipe.

app.py

@@ -0,0 +1,89 @@
+from __future__ import annotations
+
+import os
+
+import gradio as gr
+
+from pyharp import *
+from gradio_client import Client, handle_file
+
+
+_BACKEND_SPACE = "Soul-AILab/SoulX-Singer"
+_BACKEND_API_NAME = "/synthesis_function"
+_BACKEND_TOKEN_ENV = "HF_TOKEN"
+_client = None
+
+
+def _backend_client():
+    # Lazily create and cache one warm connection to the backend Space.
+    global _client
+    if _client is None:
+        _token = os.environ.get(_BACKEND_TOKEN_ENV) or None
+        _client = Client(_BACKEND_SPACE, hf_token=_token)
+    return _client
+
+
+model_card = ModelCard(
+    name="SoulX-Singer",
+    description="SoulX-Singer is a high-fidelity, zero-shot singing voice synthesis model that enables users to generate realistic singing voices for unseen singers. It supports melody-conditioned (F0 contour) and score-conditioned (MIDI notes) control for precise pitch, rhythm, and expression.",
+    author="Soul-AILab",
+    tags=["text-to-audio", "music", "singing-voice-synthesis", "svs", "zero-shot", "text-to-speech", "en", "zh"],
+)
+
+
+def process_fn(prompt_audio, target_audio, control, auto_shift, pitch_shift, seed, prompt_lyric_lang, target_lyric_lang, prompt_vocal_sep, target_vocal_sep):
+    _raw = _backend_client().predict(
+        handle_file(prompt_audio),
+        handle_file(target_audio),
+        None,
+        None,
+        control,
+        auto_shift,
+        pitch_shift,
+        seed,
+        prompt_lyric_lang,
+        target_lyric_lang,
+        prompt_vocal_sep,
+        target_vocal_sep,
+        api_name="/synthesis_function",
+    )
+    _values = list(_raw) if isinstance(_raw, (list, tuple)) else [_raw]
+    _detail = " | ".join(str(_v) for _v in _values if isinstance(_v, str) and _v.strip())
+    _out_generated_audio = _values[0] if len(_values) > 0 else None
+    if not _out_generated_audio:
+        raise gr.Error(_detail or "The backend Space returned no 'generated_audio' output. Check the backend Space's logs; if it uses ZeroGPU it may need a moment to warm up.")
+    _out_processed_prompt_metadata = _values[1] if len(_values) > 1 else None
+    if not _out_processed_prompt_metadata:
+        raise gr.Error(_detail or "The backend Space returned no 'processed_prompt_metadata' output. Check the backend Space's logs; if it uses ZeroGPU it may need a moment to warm up.")
+    _out_processed_target_metadata = _values[2] if len(_values) > 2 else None
+    if not _out_processed_target_metadata:
+        raise gr.Error(_detail or "The backend Space returned no 'processed_target_metadata' output. Check the backend Space's logs; if it uses ZeroGPU it may need a moment to warm up.")
+    return _out_generated_audio, _out_processed_prompt_metadata, _out_processed_target_metadata
+
+
+with gr.Blocks() as demo:
+    input_components = [
+        gr.Audio(type="filepath", label="Prompt audio (reference voice), max 30s").set_info("Upload an audio file (max 30 seconds) to provide the reference voice for synthesis."),
+        gr.Audio(type="filepath", label="Target audio (melody / lyrics source), max 60s").set_info("Upload an audio file (max 60 seconds) to provide the melody or lyrics source."),
+        gr.Dropdown(choices=["melody", "score"], value="melody", label="Control type", info="Choose the control type for synthesis: 'melody' for F0 contour or 'score' for MIDI notes."),
+        gr.Checkbox(value=True, label="Auto pitch shift", info="Automatically adjust pitch shift to match the target audio's range."),
+        gr.Slider(minimum=-12, maximum=12, step=1, value=0, label="Pitch shift (semitones)", info="Manually adjust the pitch shift in semitones. Auto pitch shift will be ignored if a non-zero value is set."),
+        gr.Number(value=12306, label="Seed", info="Random seed for reproducibility."),
+        gr.Dropdown(choices=["English", "Chinese"], value="English", label="Prompt lyric language", info="Select the language of the lyrics in the prompt audio."),
+        gr.Dropdown(choices=["English", "Chinese"], value="English", label="Target lyric language", info="Select the language of the lyrics in the target audio."),
+        gr.Checkbox(value=False, label="Prompt vocal separation", info="Enable vocal separation for the prompt audio if it contains accompaniment."),
+        gr.Checkbox(value=True, label="Target vocal separation", info="Enable vocal separation for the target audio if it contains accompaniment."),
+    ]
+    output_components = [
+        gr.Audio(type="filepath", label="Generated audio").set_info("The synthesized singing voice."),
+        gr.File(type="filepath", label="Processed Prompt Metadata", file_types=[".mid", ".midi"]).set_info("Metadata file generated from the prompt audio."),
+        gr.File(type="filepath", label="Processed Target Metadata", file_types=[".mid", ".midi"]).set_info("Metadata file generated from the target audio."),
+    ]
+    build_endpoint(
+        model_card=model_card,
+        input_components=input_components,
+        output_components=output_components,
+        process_fn=process_fn,
+    )
+
+demo.queue().launch(share=True, show_error=False, pwa=True)

requirements.txt

@@ -0,0 +1,3 @@
+git+https://github.com/TEAMuP-dev/pyharp.git@v0.3.0
+gradio>=4.0
+gradio_client