v2 added

app/matcher.py

@@ -19,6 +19,6 @@ def sequence_match(a, b):
 
 
 def match(original, transcription):
-    sequence = sequence_match(original, transcription)
-    phonetic = phonetic_match(original, transcription)
+    sequence = sequence_match(original.lower(), transcription.lower())
+    phonetic = phonetic_match(original.lower(), transcription.lower())
     return sequence, phonetic

app/passing.py

@@ -5,7 +5,7 @@ def normalize_euclidean(euclidean, max_value):
     """
     return max(0, 100 - (euclidean / max_value) * 100)
 
-def calculate_passing(sequence, phonetic, cosine, euclidean, passing_threshold=60, euclidean_max=200):
+def calculate_passing(sequence, phonetic, cosine=0, euclidean=0, passing_threshold=50, euclidean_max=200):
     # Normalize sequence and phonetic to 0-100 scale
     sequence_normalized = sequence * 100
     phonetic_normalized = phonetic * 100
@@ -17,10 +17,10 @@ def calculate_passing(sequence, phonetic, cosine, euclidean, passing_threshold=6
     
     # Calculate the weighted average
     weights = {
-        'sequence': 0.35,
-        'phonetic': 0.35,
-        'cosine': 0.10,
-        'euclidean': 0.20
+        'sequence': 0.50,
+        'phonetic': 0.50,
+        'cosine': 0,
+        'euclidean': 0
     }
     
     weighted_score = (

app/routers/V1/voice/voice_router.py

@@ -55,10 +55,7 @@ async def transcribe_audio(
             text = get_transcription(filename_recorded)
             text = clean_transcription(text)
             sequence, phonetic = match(matcher_text, text)
-            Euclidean, Cosine = mfcc_similarty_check(
-                filename_original, filename_recorded
-            )
-            weighted_score, is_passing = calculate_passing(sequence, phonetic, Cosine, Euclidean)
+            weighted_score, is_passing = calculate_passing(sequence, phonetic)
             return JSONResponse(
                 {
                     "transcription": text,

app/transcriber.py

@@ -6,7 +6,8 @@ from datasets import load_dataset
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
-model_id = "MothersTongue/mother_tongue_model"
+model_id = "openai/whisper-large-v3"
+# model_id = "MothersTongue/mother_tongue_model"
 
 model = AutoModelForSpeechSeq2Seq.from_pretrained(
     model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True