Update app.py

app.py

@@ -1,22 +1,31 @@
 import os
-# from huggingface_hub import snapshot_download
-from modelscope.hub.snapshot_download import snapshot_download
+import spaces
+
+REPO_TYPE = "hf" 
+if REPO_TYPE not in ["hf", "ms"]:
+    raise ValueError("REPO_TYPE must be either 'hf' for Hugging Face or 'ms' for ModelScope.")
+
+if REPO_TYPE == "hf":
+    from huggingface_hub import snapshot_download
+else:
+    from modelscope.hub.snapshot_download import snapshot_download
+
 
 
 # 1. 定义本地路径和远程仓库ID
 FUN_ASR_NANO_LOCAL_PATH = "./Fun-ASR/model"
-FUN_ASR_NANO_REPO_ID = "FunAudioLLM/Fun-ASR-Nano-2512"
 SENSE_VOICE_SMALL_LOCAL_PATH = "./Fun-ASR/model/SenseVoiceSmall"
-# SENSE_VOICE_SMALL_REPO_ID = "FunAudioLLM/SenseVoiceSmall"
-# REPO_TYPE = "hf"  # "hf" for Hugging Face, "ms" for ModelScope
-SENSE_VOICE_SMALL_REPO_ID = "iic/SenseVoiceSmall"
-REPO_TYPE = "ms"  
+
+
+if REPO_TYPE == "ms":
+    FUN_ASR_NANO_REPO_ID = "FunAudioLLM/Fun-ASR-Nano-2512"
+    SENSE_VOICE_SMALL_REPO_ID = "iic/SenseVoiceSmall"
+else:
+    FUN_ASR_NANO_REPO_ID = "FunAudioLLM/Fun-ASR-Nano-2512"
+    SENSE_VOICE_SMALL_REPO_ID = "FunAudioLLM/SenseVoiceSmall"
 
 # 2. 检查本地是否存在，不存在则下载
 if not os.path.exists(FUN_ASR_NANO_LOCAL_PATH):
-    from modelscope import HubApi
-    api= HubApi()
-    api.login(os.getenv("MODELSCOPE_TOKEN"))
     print(f"正在下载模型 Fun-ASR-Nano 到 {FUN_ASR_NANO_LOCAL_PATH} ...")
     snapshot_download(
         repo_id=FUN_ASR_NANO_REPO_ID,
@@ -60,14 +69,10 @@ from funasr.utils.postprocess_utils import rich_transcription_postprocess
 # Model configurations for Hugging Face deployment
 FUN_ASR_NANO_MODEL_PATH_LIST = [
     "Fun-ASR/model", # local path, ms
-    "FunAudioLLM/fun-asr-nano", # huggingface model repo, hf
-    "FunAudioLLM/fun-asr-nano" # ModelScope model repo, ms
 ]
 
 SENSEVOICE_MODEL_PATH_LIST = [
     "Fun-ASR/model/SenseVoiceSmall", # local path together with this hf space
-    "FunAudioLLM/SenseVoiceSmall", # huggingface model repo
-    "iic/SenseVoiceSmall" # ModelScope model repo
 ]
 
 class LogCapture(io.StringIO):
@@ -401,6 +406,7 @@ def get_model_options(pipeline_type):
 # Dictionary to store loaded models
 loaded_models = {}
 
+@spaces.GPU()
 def transcribe_audio(audio_input, audio_url, proxy_url, proxy_username, proxy_password, pipeline_type, model_id, download_method, start_time=None, end_time=None, verbose=False):
     """
     Transcribes audio from a given source using SenseVoice.
@@ -485,7 +491,7 @@ def transcribe_audio(audio_input, audio_url, proxy_url, proxy_username, proxy_pa
                     vad_kwargs={"max_single_segment_time": 30000},
                     device=device,
                     disable_update=True,
-                    hub=REPO_TYPE,
+                    hub='ms',
                 )
             elif pipeline_type == "sensevoice":
                 model = AutoModel(
@@ -495,7 +501,7 @@ def transcribe_audio(audio_input, audio_url, proxy_url, proxy_username, proxy_pa
                     vad_kwargs={"max_single_segment_time": 30000},
                     device=device,
                     disable_update=True,
-                    hub=REPO_TYPE,
+                    hub='ms',
                 )
             else:
                 error_msg = "Invalid pipeline type. Only 'sensevoice' is supported."
@@ -650,20 +656,7 @@ with gr.Blocks() as iface:
     2. **Select Pipeline Type**: Choose from available pipelines:
        - **Fun-ASR-Nano** (default) - Large language model based ASR model
        - **SenseVoice** - CTC-based based ASR model with VAD
-
-    3. **Available Model Options**:
-
-       **For Fun-ASR-Nano:**
-       - `Fun-ASR/model` (local path, default)
-       - `FunAudioLLM/fun-asr-nano` (HuggingFace)
-       - `FunAudioLLM/fun-asr-nano` (ModelScope)
-
-       **For SenseVoice:**
-       - `Fun-ASR/model/SenseVoiceSmall` (local path, default for this pipeline)
-       - `FunAudioLLM/SenseVoiceSmall` (HuggingFace)
-       - `iic/SenseVoiceSmall` (ModelScope)
-
-    4. **Local Testing**: For development, you can use local paths as shown above
+    3. **Local Testing**: For development, you can use local paths as shown above
 
     Supported languages: 
         - Fun-ASR-Nano: more than 50 languages and Chinese dialects.