update

README.md

@@ -1,6 +1,6 @@
 ---
 title: ConST Speech2text Translator
-emoji: 🎙️
+emoji: 🎙️📖
 colorFrom: yellow
 colorTo: purple
 sdk: gradio

app.py

@@ -29,13 +29,13 @@ LANGUAGE_CODES = {
 
 LANG_GEN_SETUPS = {
     "de": {"beam": 10, "lenpen": 0.7},
-    "es": {"beam": 10, "lenpen": 0.7},
-    "fr": {"beam": 10, "lenpen": 0.7},
-    "it": {"beam": 10, "lenpen": 0.7},
-    "nl": {"beam": 10, "lenpen": 0.7},
-    "pt": {"beam": 10, "lenpen": 0.7},
-    "ro": {"beam": 10, "lenpen": 0.7},
-    "ru": {"beam": 10, "lenpen": 0.1},
+    "es": {"beam": 10, "lenpen": 0.1},
+    "fr": {"beam": 10, "lenpen": 1.0},
+    "it": {"beam": 10, "lenpen": 0.5},
+    "nl": {"beam": 10, "lenpen": 0.4},
+    "pt": {"beam": 10, "lenpen": 0.9},
+    "ro": {"beam": 10, "lenpen": 1.0},
+    "ru": {"beam": 10, "lenpen": 0.3},
 }
 
 os.system("git clone https://github.com/ReneeYe/ConST")
@@ -111,9 +111,10 @@ def generate(model_path):
     return output.read().strip()
 
 
-def remove_temp_files():
+def remove_temp_files(audio_file):
     os.remove("temp.txt")
     os.remove("data/test_case.tsv")
+    os.remove(f"data/{audio_file}")
 
 
 def run(audio_file, language):
@@ -123,7 +124,7 @@ def run(audio_file, language):
         get_vocab_and_yaml(language)
         model_path = get_model(language)
         generated_output = generate(model_path)
-        remove_temp_files()
+        remove_temp_files(converted_audio_file)
         return generated_output
     except:
         traceback.print_exc()
@@ -146,7 +147,7 @@ iface = gr.Interface(
     examples=[['case1.wav', "German"],['case2.wav', "German"], ['case3.wav', "German"]],
     title="ConST: an end-to-end speech translator",
     description="End-to-end Speech Translation Live Demo for English to eight European languages.",
-    article="ConST is an end-to-end speech translation model (see paper <a href='https://arxiv.org/abs/2205.02444', target='_blank'>here</a>). "
+    article="ConST is an end-to-end speech translation model (see paper at https://arxiv.org/abs/2205.02444 ). "
             "Its motivation is to use contrastive learning method to learn similar representations for semantically similar speech and text.",
     theme="seafoam",
     layout='vertical',