write files to /data directory becuase ./ directory is not allowed to write in HF spaces

ASR_Server.py

@@ -14,81 +14,12 @@ job_status = {
     "message": "No job running"
 }
 csv_path = "test.csv" 
+output_dir="/data"
 df = pd.read_csv(csv_path)
 print(f"CSV Loaded with {len(df)} rows")
 
-# def generateTranscript(ASR_model, dataset, csv_path, output_dir="./"):
-#     import os
-#     import time
-#     import pandas as pd
-#     import librosa
-#     import tqdm
-#     from transformers import pipeline
-
-#     os.makedirs(output_dir, exist_ok=True)
-#     # output_csv_path = os.path.join(output_dir, f"test_with_{ASR_model.replace('/', '_')}.csv")
-#     output_csv_path = os.path.join(output_dir, f"test_with_{ASR_model}.csv")
-
-#     if os.path.exists(output_csv_path):
-#         print(f"Transcript already exists for model {ASR_model}. Skipping transcription.")
-#         return
-
-#     # Load metadata CSV
-#     df = pd.read_csv(csv_path)
-#     print(f"CSV Loaded with {len(df)} rows")
-
-#     # Prepare
-#     df[df.columns[0]] = df[df.columns[0]].str.strip().str.lower()
-#     filename_column = df.columns[0]
-#     transcripts = []
-#     rtfx_score = []
-
-#     # Load ASR model
-#     pipe = pipeline("automatic-speech-recognition", model=ASR_model)
-
-#     # Create a map of dataset samples by file name (assumes filename is in dataset)
-#     dataset_map = {
-#         sample["audio"]["path"].split("/")[-1].lower(): sample for sample in dataset
-#     }
-
-#     for idx, row in tqdm.tqdm(df.iterrows(), total=len(df)):
-#         filename = row[filename_column].strip().lower() + ".wav"
-#         if filename in dataset_map:
-#             sample = dataset_map[filename]
-#             try:
-#                 audio_array = sample["audio"]["array"]
-#                 sample_rate = sample["audio"]["sampling_rate"]
-
-#                 start_time = time.time()
-#                 result = pipe({"array": audio_array, "sampling_rate": sample_rate})
-#                 end_time = time.time()
-
-#                 transcript = result["text"]
-#                 duration = librosa.get_duration(y=audio_array, sr=sample_rate)
-#                 rtfx = (end_time - start_time) / duration if duration > 0 else 0
-
-#                 transcripts.append(transcript)
-#                 rtfx_score.append(rtfx)
-
-#                 print(f"✅ {filename}: RTFX = {rtfx:.2f}")
-
-#             except Exception as e:
-#                 print(f"❌ Error with {filename}: {e}")
-#                 transcripts.append("")
-#                 rtfx_score.append(0)
-#         else:
-#             print(f"⚠️ File not in dataset: {filename}")
-#             transcripts.append("")
-#             rtfx_score.append(0)
-
-#     # Append to original DataFrame
-#     df['transcript'] = transcripts
-#     df['rtfx'] = rtfx_score
-
-#     df.to_csv(output_csv_path, index=False)
-#     print(f"✅ Transcripts saved to {output_csv_path}")
-
-def generateTranscript(ASR_model, csv_path, output_dir="./"):
+
+def generateTranscript(ASR_model):
     import os
     import time
     import tqdm
@@ -109,7 +40,7 @@ def generateTranscript(ASR_model, csv_path, output_dir="./"):
     # dataset = dataset.with_format("python", decode_audio=False)
     dataset = dataset.cast_column("audio", Audio(decode=False))
 
-    output_csv_path = os.path.join("./", f"test_with_{ASR_model}.csv")
+    output_csv_path = os.path.join(output_dir, f"test_with_{ASR_model}.csv")
     # Check if transcript already exists
     if os.path.exists(output_csv_path):
         print(f"Transcript already exists for model {ASR_model}. Skipping transcription.")
@@ -122,9 +53,17 @@ def generateTranscript(ASR_model, csv_path, output_dir="./"):
     total = len(df)
     job_status["total"] = total
 
+    import torch
+    # Check if GPU is available
+    if torch.cuda.is_available():
+        device = 0
+        print("Device set to use GPU")
+    else:
+        device = -1
+        print("Device set to use CPU")  
+    
     # Initialize ASR pipeline
-    pipe = pipeline("automatic-speech-recognition", model=ASR_model, device=-1)
-    # print("Device set to use CPU")
+    pipe = pipeline("automatic-speech-recognition", model=ASR_model, device=device)
 
     # Column with filenames in the CSV
     filename_column = df.columns[0]
@@ -175,6 +114,7 @@ def generateTranscript(ASR_model, csv_path, output_dir="./"):
 
         job_status["completed"] = idx + 1
         job_status["message"] = f"Processing {idx + 1}/{total}"
+        job_status["%_completed"] = {idx + 1}*100/{total}
 
     # Save results
     df["transcript"] = transcripts
@@ -182,11 +122,11 @@ def generateTranscript(ASR_model, csv_path, output_dir="./"):
 
     os.makedirs(output_dir, exist_ok=True)
     # Create the directory if it doesn't exist
-    output_dir = os.path.dirname(os.path.join(output_dir, f"test_with_{ASR_model}.csv")) # Get the directory path
-    if not os.path.exists(output_dir):  # Check if directory exists
-        os.makedirs(output_dir)  # Create directory if it doesn't exist
-        print(f"Created directory: {output_dir}")
-        
+    csv_output_dir = os.path.dirname(output_csv_path) # Get the directory path
+    if not os.path.exists(csv_output_dir):  # Check if directory exists
+        os.makedirs(csv_output_dir)  # Create directory if it doesn't exist
+        print(f"Created directory: {csv_output_dir}")
+    
     df.to_csv(output_csv_path, index=False)
     job_status["running"] = False
     job_status["message"] = "Transcription completed."
@@ -221,7 +161,7 @@ def asr_models():
         "ESPnet"
     ]
     def background_job():
-        generateTranscript("openai/whisper-base", csv_path, output_dir="./")
+        generateTranscript("openai/whisper-base")
 
     # Start the background job in a separate thread
     threading.Thread(target=background_job).start()