Add Files -Update New

alm_pipeline.py

@@ -4,14 +4,18 @@ import warnings
 import whisper
 import librosa
 import numpy as np
+import tensorflow as tf
 import tensorflow_hub as hub
+import csv
 
-# Reduce TensorFlow log noise and avoid attempting GPU / oneDNN on CPU-only envs
+# Reduce TensorFlow log noise and avoid attempting GPU / oneDNN on CPU-only envs.
+# NOTE: These env vars must be set before TensorFlow fully initializes; setting them
+# here greatly reduces, but may not completely remove, startup logs on some platforms.
 os.environ.setdefault("TF_CPP_MIN_LOG_LEVEL", "2")  # hide INFO/WARNING logs
 os.environ.setdefault("TF_ENABLE_ONEDNN_OPTS", "0")  # disable oneDNN custom ops
 os.environ.setdefault("CUDA_VISIBLE_DEVICES", "-1")  # don't try to use CUDA GPUs
 
-# Suppress specific library warnings that are expected in this setup
+# Suppress specific library warnings that are expected in this setup.
 warnings.filterwarnings(
     "ignore",
     category=UserWarning,
@@ -28,7 +32,14 @@ asr_model = whisper.load_model("small")
 
 # Load YAMNet for sound classification
 yamnet = hub.load("https://tfhub.dev/google/yamnet/1")
-class_map = yamnet.class_map_path().numpy()
+class_map_path = yamnet.class_map_path().numpy()
+if isinstance(class_map_path, bytes):
+    class_map_path = class_map_path.decode("utf-8")
+
+# Parse YAMNet class map CSV to get human-readable labels
+with tf.io.gfile.GFile(class_map_path) as f:
+    reader = csv.DictReader(f)
+    yamnet_labels = [row["display_name"] for row in reader]
 
 # Simple Emotion Estimator (from YAMNet embedding)
 def estimate_emotion(activation):
@@ -53,13 +64,8 @@ def detect_sound(audio):
     scores, embeddings, _ = yamnet(waveform)
     mean_scores = np.mean(scores.numpy(), axis=0)
     top_idx = int(np.argmax(mean_scores))
-    # class_map may contain integers or byte strings depending on TF Hub version;
-    # convert robustly to a human-readable label.
-    label = class_map[top_idx]
-    if isinstance(label, bytes):
-        label = label.decode("utf-8")
-    else:
-        label = str(label)
+    # Look up human-readable class label from YAMNet's CSV class map
+    label = yamnet_labels[top_idx] if 0 <= top_idx < len(yamnet_labels) else "Unknown"
     return label, float(mean_scores.max())
 
 

reasoning.py

@@ -14,5 +14,6 @@ Speakers: {summary['speakers']}
 Question: {question}
 Provide a detailed reasoning-based answer using the audio cues.
 """
-    result = reasoner(prompt, max_length=200)[0]["generated_text"]
+    # Use only max_new_tokens to avoid Hugging Face warning about max_length+max_new_tokens.
+    result = reasoner(prompt, max_new_tokens=256)[0]["generated_text"]
     return result