Load in the worker, declare the format, and pin requirements

Preloads the models in the worker alone, declares 24 kHz mono, cleans the README, and pins requirements. It also imports spaces plainly, since it is always installed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

README.md

@@ -14,6 +14,8 @@ tags:
   - subcategory:text-to-speech
   - input:text
   - output:audio
+  - sample-rate:24000
+  - channels:mono
   - multilingual
 ---
 
@@ -21,7 +23,4 @@ tags:
 
 Bark is a transformer-based text-to-audio model created by Suno. Bark can generate highly realistic, multilingual speech as well as other audio - including music, background noise and simple sound effects. The model can also produce nonverbal communications like laughing, sighing and crying.
 
-- Inputs: textbox, dropdown, slider, slider
-- Outputs: audio
-
-Generated by the HARP model agent from a recipe.
+A HARP deployment. See https://github.com/TEAMuP-dev/HARP

app.py

@@ -1,21 +1,8 @@
 from __future__ import annotations
 
-import gradio as gr
-try:
-    import spaces
-except ImportError:  # 'spaces' is only provided by Hugging Face Spaces
-    import types as _types
-
-    def _gpu(*args, **kwargs):
-        if len(args) == 1 and callable(args[0]) and not kwargs:
-            return args[0]
-
-        def _decorator(func):
-            return func
-
-        return _decorator
+import spaces  # imported first, as ZeroGPU requires
 
-    spaces = _types.SimpleNamespace(GPU=_gpu)
+import gradio as gr
 
 from pyharp import *
 
@@ -39,10 +26,6 @@ except Exception:  # torch not installed / unexpected API -- nothing to patch
 from bark import generate_audio, preload_models, SAMPLE_RATE
 from scipy.io.wavfile import write as write_wav
 import tempfile
-import os
-
-# Download and load all models once
-preload_models()
 
 
 model_card = ModelCard(
@@ -52,12 +35,18 @@ model_card = ModelCard(
         "can generate highly realistic, multilingual speech as well as other "
         "audio - including music, background noise and simple sound effects. "
         "The model can also produce nonverbal communications like laughing, "
-        "sighing and crying. Repository: https://github.com/suno-ai/bark"
+        "sighing and crying. It writes 24 kHz mono audio. Repository: "
+        "https://github.com/suno-ai/bark"
     ),
     author="suno-ai",
-    tags=[Subcategory.TEXT_TO_SPEECH, 'multilingual'],
+    tags=[Subcategory.TEXT_TO_SPEECH, SampleRate(SAMPLE_RATE), Channels(1), 'multilingual'],
 )
 
+# pyharp runs process_fn in a worker, which imports this file as __mp_main__ and starts with the
+# server, so every model is downloaded and loaded there alone
+if __name__ != "__main__":
+    preload_models()
+
 
 def process_fn(text_prompt, history_prompt, text_temp, waveform_temp):
     # Checked before the GPU is reserved, so a request the model cannot take spends none of the

requirements.txt

@@ -1,13 +1,13 @@
 git+https://github.com/TEAMuP-dev/pyharp.git@v0.4.2
 git+https://github.com/suno-ai/bark.git@f4f32d4cd480dfec1c245d258174bc9bde3c2148
-boto3
+boto3==1.43.118
 encodec==0.1.1
-funcy
-huggingface-hub
-numpy
-scipy
-tokenizers
-torch
-tqdm
-transformers
+funcy==2.1
+huggingface-hub==1.33.0
+numpy==2.2.6
+scipy==1.15.3
+tokenizers==0.23.3
+torch==2.13.0
+tqdm==4.70.1
+transformers==5.20.2
 spaces>=0.51.1