[update] cli package

app.py

@@ -14,7 +14,8 @@
 # limitations under the License.
 
 import gradio as gr
-import wespeakerruntime as wespeaker
+import wespeaker
+# import wespeakerruntime as wespeaker
 # from sklearn.metrics.pairwise import cosine_similarity
 
 STYLE = """
@@ -46,8 +47,8 @@ OUTPUT_ERROR = (STYLE + """
     </div>
 """)
 
-vox_model = wespeaker.Speaker('pre_model/voxceleb_resnet34.onnx')
-cnc_model = wespeaker.Speaker('pre_model/cnceleb_resnet34.onnx')
+cn_model = wespeaker.load_model("chinese")
+en_model = wespeaker.load_model("english")
 
 
 def speaker_verification(audio_path1, audio_path2, lang='CN'):
@@ -55,16 +56,17 @@ def speaker_verification(audio_path1, audio_path2, lang='CN'):
         output = OUTPUT_ERROR.format('Please enter two audios')
         return output
     if lang == 'EN':
-        model = vox_model
+        model = cn_model
     elif lang == 'CN':
-        model = cnc_model
+        model = en_model
     else:
         output = OUTPUT_ERROR.format('Please select a language')
         return output
-    emb1 = model.extract_embedding(audio_path1)
-    emb2 = model.extract_embedding(audio_path2)
-    cos_score = model.compute_cosine_score(emb1[0], emb2[0])
-    cos_score = (cos_score + 1) / 2.0
+    # emb1 = model.extract_embedding(audio_path1)
+    # emb2 = model.extract_embedding(audio_path2)
+    # cos_score = model.compute_cosine_score(emb1[0], emb2[0])
+    # cos_score = (cos_score + 1) / 2.0
+    cos_score = wespeaker.compute_similarity(audio_path1, audio_path2)
 
     if cos_score >= 0.70:
         output = OUTPUT_OK.format(cos_score * 100)

requirements.txt

@@ -1,3 +1,3 @@
 gradio
-wespeakerruntime
+wespeaker @ git+https://github.com/wenet-e2e/wespeaker.git
 scikit-learn